WO2025088220A1 - Artificial intelligence-based model for patient-individual planning of a refractive eye surgery - Google Patents

Abstract

The invention relates to a computer program comprising an artificial intelligence-based model, the computer program being designed to cause a computer upon execution of the computer program by said computer and by means of the model to determine information characterizing a treatment plan for a refractive eye surgery, the determination being based on information characterizing a patient and information characterizing a refractive error of an eye of the patient that should be corrected by means of the refractive eye surgery.

Description

ARTIFICIAL INTELLIGENCE-BASED MODEL FOR PATIENT-INDIVIDUAL PLANNING OF REFRACTIVE EYE SURGERY

The present disclosure relates to a computer program comprising an artificial intelligence-based model, an apparatus for data processing with the model, a method for training the model, and a computer-readable medium with the model.

The disclosed model can be used for patient-specific planning of laser refractive eye surgery for one of a patient's eyes. Refractive eye surgery, or refractive surgery, encompasses eye operations that alter the overall refractive power of the eye. One goal may be to eliminate the need for conventional optical corrections such as glasses or contact lenses, or at least to reduce their required power. Ophthalmology knows several surgical methods for correcting ametropia or refractive errors through refractive eye surgery.

A fundamental distinction is made between laser procedures and implants, particularly intraocular lenses, which are inserted into the optical beam path. The present disclosure relates to the correction of refractive errors using a laser procedure and/or an implant.

Laser procedures use a laser to alter the refractive power of the eye to correct the visual defect or ametropia. This can be done indirectly, for example, by treating the cornea, and/or directly, for example, by creating an incision for inserting an intraocular lens (IOL).

One laser procedure is LASIK (laser in situ keratomileusis). This procedure uses a microkeratome (corneal planer) or a femtosecond laser (so-called Femto-LASIK) cuts a thin lamella (approximately 8 to 9.5 mm in diameter and between 100 and 160 pm thick) into the cornea. This lamella is opened, and the actual laser treatment takes place on the underlying tissue.

Other laser procedures include photorefractive keratectomy (PRK), laser epithelial keratomileusis (LASEK), and epithelial laser in situ keratomileusis (EpiLASIK). In these procedures, the tissue is removed from the surface of the cornea. This is why they are also referred to as surface ablation. In all three methods, the epithelium is first removed from a sufficiently large, central area of the cornea, and then the corneal surface is treated with the laser. The procedures differ in how the epithelium is removed and what happens to it after treatment. In PRK, the epithelium is scraped off using a surgical instrument and is not reused. The LASEK and EpiLASIK procedures use the epithelium as a natural wound dressing after treatment. In LASEK, the epithelium is loosened with alcohol and pushed aside with a suitable instrument. In EpiLASIK, however, it is lifted with a blunt corneal plane similar to a microkeratome, forming a kind of epithelial flap. In all three procedures, the epithelium must regenerate after treatment.

Another well-known procedure is femtosecond lenticule extraction. As with the excimer laser procedures PRK and LASIK, the refractive error is corrected by changing the curvature of the cornea. However, unlike the aforementioned procedures, this is not achieved by vaporizing corneal tissue. A femtosecond laser is used to cut a so-called lenticule within the cornea. This lens-shaped piece of tissue is then removed, and the resulting change in the curvature of the cornea corrects the refractive error. Depending on how this lenticule is removed, a distinction is made between two procedures. In the FLEx (Femtosecond Lenticle Extraction) method, not only the lenticule is cut during the laser treatment, but also an overlying lamella (flap). This flap is then opened, allowing the lenticule to be removed. In the second method, called SmILE or SMILE (small incision lenticle extraction), the laser does not create a complete flap, but rather only a small, peripheral incision through which the doctor can remove the lenticule. Femtosecond lenticule extraction is generally indicated for the correction of myopia up to -10 diopters and astigmatism up to 3 diopters.

In the aforementioned refractive eye surgery using a laser, patient-individual (treatment) planning is usually carried out.

When planning refractive eye surgery, the practitioner derives appropriate treatment parameters (e.g., target, laser set, optical zone, laser energy, laser spot parameters, etc.) based on patient data (e.g., gender, age, etc.) and diagnostic data (e.g., manifest refraction, pachymetry, etc.). The patient data and diagnostic data can include numerical values (e.g., measured values), text data, videos, and/or image data (e.g., diagnostic images).

The planning parameters, i.e. the patient data and diagnostic data, are entered by the practitioner into the planning software of the laser or the associated workstation.

In this context, reference is made to US 2020/155351 A1, which concerns the development of treatment recommendations for topography-based surgical excimer laser procedures.

In conventional planning for refractive eye surgery, the planning parameters are selected and entered manually. This can result in the overall refractive procedure being slowed down due to the large number of inputs. Furthermore, input errors can occur. Advice on selecting appropriate planning parameters or the appropriate treatment type for the individual patient, based on recommendations from other physicians or previously entered data, is also not available. US 2018/161098 A1 relates to providing personalized refractive surgery recommendations for eye patients. A computer system accesses preoperative ocular feature data for the patient's eyes. The preoperative ocular feature data originates from preoperative diagnostic procedures performed on the patient. The computer system accesses demographic data of the patient. The computer system inputs the preoperative ocular feature data and the demographic data into a predictive model in the system memory. The predictive model was calculated from patient data of other patients who have previously undergone refractive surgery. For each other patient, this patient data includes preoperative ocular feature data, demographic data, a refractive surgery type, and/or a postoperative uncorrected visual acuity value. The predictive model transforms the preoperative ocular feature data, the demographic data, and other patient data using linear regression. Using linear regression, the predictive model predicts one or more postoperative uncorrected visual acuity maps for the patient. Each of the one or more postoperative uncorrected visual acuity maps (UCVA) is derived for a corresponding refractive surgery type based on the patient's preoperative ocular characteristics and demographic data, taking into account the other patient data. Each of the uncorrected visual acuity maps predicts the postoperative uncorrected visual acuity values for the patient in one or more postoperative periods for the corresponding refractive surgery type. Using linear regression, the predictive model assigns the patient to a selected refractive surgery type based on the predicted postoperative uncorrected visual acuity values in the one or more uncorrected visual acuity maps for the patient. The predictive model recommends the selected refractive surgery type as the refractive surgery recommendation for the patient. The predictive model also determines a set of surgical parameters for the selected refractive surgery type. The predictive model can surgical parameters together with the selected refractive surgery type in a recommendation.

It is an object of the present disclosure to enrich the prior art outlined above.

According to the disclosure, this object is achieved by the subject matter of the independent claims, with optional further developments thereof being specified in the subordinate claims and the subclaims.

The task is solved by a computer program comprising a model based on artificial intelligence.

The computer program is designed so that, when the computer program is executed by a computer, the computer, by means of the model, causes the computer to determine information characterizing a treatment plan for the refractive eye surgery based on information characterizing a patient and information characterizing a refractive error of an eye of the patient that is to be treated by means of the refractive eye surgery.

The computer program includes a recommendation system. The recommendation system is designed to prompt the computer, when the computer executes the computer program, to use the recommendation system to identify at least one additional patient from a plurality of patients based on the information identifying the patient and/or the information identifying the refractive error of the patient's eye that is to be treated by refractive eye surgery, and to determine additional information identifying a treatment plan for refractive eye surgery performed on the at least one identified patient. In other words, an artificial intelligence-based model is provided in the form of or as part of a computer program. This means that the artificial intelligence-based model is designed to be used by a computer as part of or in the form of a computer program. The model is trained in such a way that, based on input data, it automatically generates output data, which in turn corresponds to a partially or fully automated patient-specific planning of a refractive eye surgery for a patient's eye. The input data for the model consists of information characterizing the patient and information characterizing a refractive error in the patient's eye that is to be treated, or optionally corrected, by means of the refractive eye surgery. This input data is processed by the model when the computer executes the computer program in such a way that the output data is generated, which includes information characterizing a treatment plan for the refractive eye surgery.

In addition, a recommendation system is provided in the form of or as part of the computer program. This recommendation system uses as input data, just like the model, at least the information characterizing the patient and/or the information characterizing a visual impairment of the patient's eye that is to be treated, optionally corrected, by means of refractive eye surgery. Based on this input data, the recommendation system identifies or determines a further patient from a large number of further patients based on predetermined deterministic rules and/or using a further system based on artificial intelligence. The large number of further patients can be made available to the recommendation system, for example, in the form of a data set or a database. It is conceivable that the database for each of the further patients includes information regarding each input data used by the recommendation system. This means that the database can contain for each of the further patients at least one (associated) (respective) additional patient information and/or the information that identifies the refractive error of the eye of the (respective) additional patient that was treated by means of the refractive eye surgery. For each of the additional patients, the additional information that identifies the treatment plan for the refractive eye surgery performed on the (respective) additional patient can be retained, e.g., as part of the data set or database. Based on this, the recommendation system can determine the additional information that identifies the treatment plan for the refractive eye surgery performed on the identified patient.

The model can thus enable patient-specific planning of a refractive eye surgery for a patient's eye, e.g., using a laser and/or an implant.

A computer program can be understood as a sequence of instructions that complies with the rules of a specific programming language in order to process or solve certain functions, tasks or problems, in this case a patient-specific planning of a refractive eye operation for one eye of a patient, with the help of a computer.

A computer program is part of a computer's software. It can be present (e.g., on a data storage device) as a computer-executable program file, often in so-called machine code, which can be loaded into the computer's memory for execution. The computer program is processed and thus executed as a sequence of machine or processor instructions by the computer's processor(s).

For the definition of refractive eye surgery, reference is made to the above explanations, particularly regarding laser procedures. Consequently, a laser may be used in eye surgery, for example, a femtosecond laser and/or an excimer laser. It is conceivable that The artificial intelligence-based model is specifically tailored to the laser, meaning that during training (see details below), the model is also fed information that characterizes the laser. This could, for example, be nomograms.

In this context, an artificial intelligence-based model can be understood as a model created using machine learning. Machine learning can be understood as the "artificial" generation of knowledge from experience. An artificial system or model learns from one or more examples and can generalize them after the learning phase or training is complete. To do this, machine learning algorithms build a statistical model based on training data, which can optionally be tested against test data. This means that the examples are not simply memorized, but patterns and regularities are recognized in the training data. This allows the model to evaluate even unknown data (so-called learning transfer).

The computer program described above offers a number of advantages. Among other things, the computer program enables (partially or fully) automated, yet patient-specific planning of refractive eye surgery through its artificial intelligence-based model (Kl model), which draws on expert knowledge. Patient-specific can be understood to mean that the Kl model can be used to plan refractive eye surgery based on patient-specific information that serves as input data for the Kl model. The use of the Kl model for application in refractive eye surgery or eye surgery is advantageous in that, as part of the learning transfer described above, it can independently determine information characterizing a treatment plan (e.g., comprising a treatment procedure and/or a setting for a laser used in the eye surgery). Unlike a comparable deterministic algorithm, the Kl model is not limited to determining the initial data on a predefined, permanently implemented rule set, but can rely on (expert) knowledge acquired during the learning or training phase when determining the output data. This enables a better representation of the conventional, manual planning processes using a computer program than conventional, deterministic algorithms. In other words, the Kl model, as a knowledge-based system, makes it possible to model and thus automate the conventional, complex planning process of an experienced surgical planner, something that cannot be represented with a deterministic algorithm. In addition, there is the option of (optionally further) training the Kl model when used in the field, i.e. when used as intended, for example based on user feedback.

The interaction of the Kl model with the recommendation system described above can enhance the effects described above.

A recommendation system (for refractive planning) can be understood as a decision support system (DSS). This can be triggered consciously or unconsciously by the user. The recommendation system can be designed to determine whether there are one or more recommendations that can be made for the current planning context. These recommendations can, for example, be determined based on historical or existing data relating to the current planning context. It is conceivable that the recommendation system includes or consists of a recommender system. A recommender system can be understood as, among other things, a KL model that provides or identifies data on one or more additional patients who exhibit a predetermined degree of similarity to the current patient. This means that various KL models and/or deterministic algorithms can be used within the DSS, including: including those that belong to the group of recommender systems (such as k-nearest neighbor etc.).

It is conceivable that the recommender system could provide patient-specific suggestions, for example, one or more pieces of information that characterize the patient's treatment plan, such as the (most suitable) treatment procedure. For this purpose, the recommender system could, for example, use (anonymized) historical patient data from a database containing the respective therapy results.

Various algorithmic variants are possible for implementing the recommender system and the similarity metric, ranging from a simple k-nearest neighbor search using the cosine similarity metric to deep learning-based methods with complex feature representations and similarity metrics. Supplementation with rule-based approaches, similar to an expert system, is also conceivable. Multimodal plan comparisons are also possible (SMILE/LASIK/PRK versus phakic IOL versus inlay plan versus LIRIC versus CXL), and each clinic or user can individually set which input parameters should be used for the similarity search and which criteria should be used for ranking the suggestions.

The output of the recommendation system can be used in various ways, with some possible uses being described below as examples and not exhaustively.

To the extent that the user performs the final review of the AI model's output based on the output of the recommendation system manually or semi-automatically, the plausibility module can provide a technical effect within the framework of so-called guided user interaction. The AI model, or more precisely how the AI model determines its output, is not visible to the user. The AI model is therefore a black box. A user who, however, is familiar with The user is entrusted with the technical task of creating a treatment plan based on the given input data using the AI model described above. He or she is now faced with the challenge of reviewing the treatment plan created automatically or using the AI model. It is conceivable that the treatment plans identified using the recommendation system will be displayed. The treatment plans can originate from other patients who were treated by the same user and/or with the same device as the (current) patient. This allows the user to at least partially review the treatment plan determined by the AI model by comparing the output of the recommendation system with the output of the AI model.

The recommendation system can thus at least implicitly provide a justification for the result of the Kl model.

It is conceivable that an optional, explicit justification for the respective result of the Kl model could also be provided to the user. As part of the recommendation of a specific treatment plan, for example, a justification could be provided that shows the advantages of a recommended treatment over other treatment types in a patient-specific and individual manner, thus transparently supporting the user in the decision-making process. For example, for a patient with a thin cornea and a large optical zone who requires a high refractive correction, an IOL or LASIK treatment could be recommended over SMILE. The different residual current thickness could be pointed out as a justification, and the difference could be clearly presented. In the further course of the process, the justification for a given recommendation could also be extended to a specific parameter selection within a treatment type. For example, in a patient with a high planned refractive correction and sufficiently thick cornea, the depth of the SMILE-Cap incision can be increased from the standard depth of the doctor / practice (e.g. 110 pm) to a depth of e.g. 140 pm, so that in case of a necessary re- correction (e.g. increased risk due to unstable refraction or high refractive correction) it is possible to perform a thin-flap LASIK at a depth of 90 pm.

However, the plausibility check of the result of the Kl model is not only possible manually by the user, but can, additionally or alternatively, be carried out (fully or partially) automatically by the plausibility check module.

The computer program can therefore comprise a plausibility check module which is designed to, when the plausibility check module is executed by the computer, cause the computer to check the plausibility of the specific information based on the specific further information by means of the plausibility check module.

The recommendation system can be designed so that, when the computer program is executed by the computer, the computer, by means of the recommendation system, causes the computer to identify that further patient from the plurality of patients for whom a therapy result achieved with the refractive eye surgery performed on this further patient meets a predetermined criterion.

In detail, collaborative filtering, among other things, may be suitable for creating a recommender system. This approach can search for one or more patients who are as similar as possible to the (active) patient according to a predetermined (application-specific) similarity metric. A ranking of the most similar cases and the selected treatment options can then be displayed and compared with the treatment suggestion generated by the AI model. This allows the user and/or the computer program itself to check the plausibility of the treatment suggestion and correct it if necessary. This can mean that if the recommendation system identifies several additional patients, they can be ranked based on their treatment outcome. This can be an absolute ranking, e.g., the remaining degree of refractive error, and/or a relative ranking, e.g., how similar is the respective treatment outcome to the treatment outcome to be achieved for the (current) patient. This can provide the advantage of not only identifying the additional patient(s) who are most similar to the (current) patient according to the selected similarity metric, but also those whose treatment outcome corresponds to a target treatment outcome.

The recommendation system can be designed so that, when the computer executes the computer program, the computer uses the recommendation system to determine a similarity value based on the information identifying the patient, and/or the information identifying the refractive error of the patient's eye that is to be treated by refractive eye surgery, as well as information identifying the other patients, respectively, and/or information identifying a refractive error of one of the other patients' eyes that was treated by the refractive eye surgery performed on these patients. The recommendation system can be designed so that, when the computer executes the computer program, the computer uses the recommendation system to identify the at least one other patient based on the determined similarity value.

In other words, several possibilities for implementing the recommendation system have already been discussed above. It is conceivable that the recommendation system uses the input from the Kl model and compares this input with a database to identify (at least one) additional patient. The similarity value can be seen as the intersection between the information used as input to the recommendation system and the information stored in the database for each additional patient. The similarity value therefore does not have to be a single value, but can can also be represented as a vector. It is conceivable that the other patients could be ranked based on the similarity value.

The recommendation system can be configured to cause the computer, when the computer executes the computer program, to determine a weighting factor using the recommendation system. The recommendation system can be configured to use the weighting factor to determine intermediate results of the recommendation system and/or to rank predetermined therapeutic methods, predetermined treatment procedures, and/or predetermined therapy-relevant information with which a therapeutic result achieved with the refractive eye surgery performed on this additional patient is likely to meet a predetermined criterion.

The intermediate results can serve the system's internal logic and can, for example, be fitted internal parameters of a neural network and/or a weighting for or against a specific rule of an expert system. Such internal weighting factors could, for example, define the strength of the impact of the planned correction on the ranking of possible treatment methods. For example, the higher the correction itself (ADpt.), the less laser correction should be preferred and the more an implantable contact lens should be preferred. Thresholds and weighting factors can be defined for this behavior. Another example could concern a user's preference regarding recommendation behavior for treatment plans. Specifically, this could concern the user's preference, for example, whether they would prefer the laser correction to be undercorrected or overcorrected in case of doubt.

Regarding the ranking of predetermined therapeutic methods: In addition to the effect of a weighting on the "internal logic" of the computer system, a weighting factor can also be part of the recommendation itself (displayed to the user). In this case, the recommendation content cannot be binary (e.g., recommendation for new iris image recording yes/no), but (multiple) in the form of an evaluation of possible variants. A recommendation could then, for example, look like this: "With treatment method 1, a visual acuity of 20/20 at near vision and 20/16 at distance vision, and 95% freedom from glasses can be achieved. With treatment method 2, at near vision...". A ranking of several or all possible treatment methods can therefore be output according to their chances of success. Such a ranking can, however, not only be based on possible treatment methods, but also on the basis of various suggested treatment parameter combinations. For example, with parameter set 1, a residual current thickness of X and 95% freedom from dysphotopic phenomena (such as starburst, halo, etc.) can be achieved, with parameter set 2 .... A ranking of several or all recommended parameter sets can then be determined and output according to their chances of success (summative scoring). The chances of success can, for example, be defined in advance by the user and/or already be stored in the program.

Information that characterizes the patient can originate from both characteristics of the eye itself and general health-related patient information. Information about the eye can include biomechanical stability and biometric characteristics, previous illnesses or treatments, as well as the use and preference of visual aids. Health-related patient information can include, for example, the patient's biological age, gender, previous illnesses, medications, lifestyle information (e.g., smoking status), origin, and ethnicity.

The information characterizing the refractive error of the patient's eye to be treated by refractive surgery may include information about a manifest refraction, a higher-order aberration present before refractive surgery, and/or pachymetry. The information identifying the treatment plan for refractive eye surgery may include a treatment procedure from a list of predetermined treatment procedures.

The list may include at least two of the following treatment procedures: femtosecond lenticule extraction (SMILE), photorefractive keratectomy (PRK), laser epithelial keratomileusis (LASEK), epithelial laser in situ keratomileusis (EpiLASIK), laser in situ keratomileusis (LASIK), and/or provision of an implant, optionally an intraocular lens (IOL) and/or a phakic intraocular lens (pOL), and/or corneal intrastromal implantation surgery (CISIS).

For the definition and description of these individual procedures, please refer to the above.

The computer program according to the disclosure can thus, among other things, use the Kl model to help select a treatment method suitable for the patient and/or to optimize the clinically achieved result.

It is conceivable that the information characterizing the treatment plan for refractive eye surgery also includes laser settings for performing the treatment procedure. This may include information about a laser set, an optical zone, a specific target, a laser energy, and/or a laser spot parameter.

The computer program according to the disclosure can thus, using the Kl model, not only select a treatment method suitable for the patient, but also determine further information corresponding to or suitable for the selected treatment method, such as the setting of the laser. A target can be understood as a desired refraction or a refraction to be achieved/desired after eye treatment. The change in refractive power can be achieved through the use of an implant (see above) and/or a laser (see above).

A laser set can be understood as a piece of information, optionally a necessary refractive change. This information can be represented by control data, whereby the control data is used by the laser, i.e., provided to it to reach the target.

It is conceivable that for this purpose (e.g. a user) sets the necessary refractive changes, e.g. in the form of sphere, cylinder and/or higher orders in diopters, on the laser in order to achieve the desired refraction/target by means of the refractive procedure. If, for example, SMILE is selected as the treatment method, software can be provided that translates this entered refractive change into a lenticule geometry (i.e. radius of the lenticule incision). Further lenticule parameters (e.g. minimum thickness, OZ, depth of the lenticule in the cornea, size and/or position of the incision for lenticule extraction, a buffer or transition zone, and/or a clearance) can be specified by the user in addition to the refractive change (or the laser set). This results in the final lenticule geometry. If, for example, PRK, Trans-PRK, or LASIK is chosen as the treatment procedure, software can be provided that translates the refractive change into an ablation profile that the (excimer) laser ablates from the corneal surface (e.g., photoablation, vaporization without long-range heating of the tissue). For LASIK, additional flap parameters (e.g., flap thickness, hinge position, etc.) can be entered (e.g., by the user) in addition to the laser setting. Input by the user is only one example of possible implementation, and input of the laser set can also be semi-automated or fully automated, optionally controlled by the computer program according to the disclosure. The optical zone can be understood as an optically active area of a lenticule, flap or ablation profile to be cut.

Laser energy can be understood as the energy per laser pulse emitted by the laser during treatment of the eye, i.e., one pulse energy. Such energy can vary depending on the treatment method (for example, in the SMILE method, in a range of 115 to 150 nJ). When using an excimer laser, pulse energies between 1.04 and 1.08 mJ can occur for a Gaussian spot with a diameter of 0.7 mm. The aforementioned numerical values serve solely to explain the concept of laser energy and do not limit the disclosure to these numerical values.

The laser spot parameter can be understood as information relating to an ellipsoid spiral scan, which in turn can include information relating to spot spacing (i.e., the spatial distance between the spots on a spiral path) and/or track spacing (i.e., the spatial distance between the spiral arms). Both of the aforementioned parameters can, for example, be in a range of 0.1 to 4 micrometers, optionally 1 to 4 micrometers. The spatial distance between the spots and the spatial distance between the spiral arms can be selected to be equal. The aforementioned numerical values serve exclusively to explain the term laser spot parameter and do not limit the disclosure to these numerical values.

Furthermore, the disclosure relates to a data processing device, wherein the data processing device is configured to determine information characterizing a treatment plan for the refractive eye surgery by means of an artificial intelligence-based model based on information characterizing a patient and information characterizing a refractive error of an eye of the patient that is to be corrected by means of a refractive eye surgery. The data processing device can be configured to determine information characterizing a treatment plan for the refractive eye surgery by means of a recommendation system based on the information identifying the patient and/or the information identifying the refractive error of the patient's eye to be treated by means of the refractive eye surgery, to identify at least one further patient from a plurality of patients and to determine further information identifying a treatment plan for a refractive eye surgery performed on the at least one identified patient.

The data processing device can be designed to check the plausibility of the specific information based on the specific further information by means of a plausibility check module.

The recommendation system may be configured to identify that additional patient from the plurality of patients for whom a therapeutic outcome achieved with the refractive eye surgery performed on that additional patient meets a predetermined criterion.

The recommendation system can be configured to determine a similarity value based on the information identifying the patient, and/or the information identifying the refractive error of the patient's eye to be treated by the refractive eye surgery, as well as information identifying the other patients, respectively, and/or information identifying a refractive error of one eye of the other patients, respectively, who were treated by the refractive eye surgery performed on these patients. The recommendation system can be configured to identify the at least one other patient based on the determined similarity value.

The recommendation system can be designed to determine a weighting factor and to use the weighting factor to determine intermediate results of the recommendation system and/or to rank predetermined therapeutic methods, predetermined treatment procedures and/or predetermined therapy-relevant information with which a therapy result, achieved with the refractive eye surgery performed on this additional patient is likely to meet a predetermined criterion.

The data processing device may comprise an input interface for receiving the information identifying the patient and/or the information identifying a refractive error of the patient's eye to be treated by means of the refractive eye surgery.

The data processing device or computer can be a local processing unit that is wirelessly and/or wiredly connected to the laser and/or is embodied as part of the laser. Additionally or alternatively, it is conceivable that parts of the data processing device are arranged remotely and in turn are wirelessly and/or wiredly connected to the local processing unit. This means that it is conceivable, for example, that the AI model is provided at least partially in a cloud, e.g., as Software as a Service (SaaS). In such a case, it is conceivable that the input data required for the AI model are transferred from the local computer to a server, for example, via the Internet. The AI model can then be executed in the cloud or by the respective cloud component, which, as part of the server, is also part of the data processing devices. In this case, the output data of the AI model determined on the server, i.e., the specific laser setting characterizing the refractive eye surgery, can be output to the local computer, again, for example, via the Internet. The server and thus the computer program comprising the Kl model can be operated, maintained and, if necessary, further developed by a software manufacturer based on input data from several local computers and thus users.

The input interface may comprise an input interface, optionally comprising a display device, which is designed to receive a user input concerning the information identifying the patient and/or the information which indicates a refractive error of the patient's eye that is to be treated by refractive eye surgery.

The input interface may comprise a data receiving interface configured to receive a data signal comprising the information identifying the patient and/or the information identifying a refractive error of the patient's eye to be treated by means of the refractive eye surgery.

In the case of a distributed system, i.e., the use of a server or cloud as described above, this can be implemented in such a way that the surgical planner or user makes their input on the local computer regarding the information characterizing the patient and/or the information characterizing a visual impairment of the patient's eye that is to be treated by refractive eye surgery (i.e., via the input interface). This information received via the user input can then be forwarded to the server (i.e., to the data interface), which enters the received information into the AI model in the manner described above. It is also conceivable that the user input only comprises part of the information characterizing the patient and/or the information characterizing a visual impairment of the patient's eye that is to be corrected by refractive eye surgery. Further information can, for example, be loaded from one or more databases based on the information contained in the user input. For example, it is conceivable that the information characterizing the patient consists of the patient's name and/or an ID. Based on the patient's or subject's name and/or ID, further information identifying the patient (e.g., gender and/or age, see above) and/or information identifying a refractive error in the patient's eye that is to be corrected by refractive surgery can be loaded from one or more databases. This allows information from multiple sources to be combined and the amount of data to be transferred to be kept to a minimum, which also can be advantageous from a security perspective regarding sensitive personal data.

The data processing device may comprise an output interface for outputting the determined information characterizing the treatment plan for the refractive eye surgery.

The output interface may comprise an output interface for outputting the specific information characterizing the treatment plan to a user of the device. This may, for example, be a display or a display device via which the information characterizing the treatment plan is visibly displayed or visualized for the user.

The output interface can, for example, be configured to output feasible parameter ranges in relation to the information determined by the AI model (with the aid of the artificial intelligence-based model in a user-selected refractive planning context). For example, it is conceivable that a minimum/maximum feasible or feasible optical zone is displayed and a warning is issued if the planning or the information determined by the computer program, or more precisely, the AI model, falls outside this range.

The output interface can, for example, be configured to output information on refractive planning quality. Planning quality can refer to the quality of the input to the AI model, such as a diagnostic input. For example, the AI model can detect that one of five acquired iris images is an outlier. This can reduce planning reliability, which in turn can be displayed to the user as feedback.

The output interface can, for example, be designed to display recommendations for refractive result optimization and suggest them to the user for adoption. This means that the user can, for example, instruct the computer program Specify specifications for determining the information that characterizes the treatment plan, such as a specific flap thickness. The computer program can then be designed to review these user specifications and determine whether a modified specification or a modified framework, such as a modified flap thickness, leads to an improved treatment outcome. If this is the case, the modified framework can be suggested to the user. The computer program can be structured so that the user can accept or reject the suggestion, thus overriding the system at any time.

The output interface may comprise a data output interface configured to output a data signal comprising the specific information identifying the treatment plan. This may, for example, be an interface to a database in which the specific information identifying the treatment plan is stored.

It is conceivable that the device is configured to determine a control signal for a laser used in the eye surgery based on the specific information characterizing the treatment plan and to output the specific control signal to the laser via the output interface for controlling the laser. It is conceivable that the laser itself can access the output control signal. This allows, at least in part, remote control of the laser to be realized.

It is conceivable that the device is designed to access a plurality of different computer-implemented models based on artificial intelligence, by means of which the data processing device is designed in each case to determine the information characterizing the treatment plan for the refractive eye surgery based on the information characterizing the patient and the information characterizing the refractive error of the patient's eye that is to be treated by means of the refractive eye surgery. One or more of the models may be selectable via the input interface of the device to determine the information characterizing the treatment plan for the refractive eye surgery.

It is conceivable that the output of several or all of these models can be checked for plausibility using the recommendation system and optionally the most plausible output can be or is selected.

It is conceivable that a user could choose between several AI models or assistants when planning or determining the laser setting. For this purpose, the input interface described above could be used, for example. The available assistants can be trained with user data and/or expert data. The assistant can therefore either be used to speed up the user's planning by being trained with its own data, or it can serve as an advisor based on recommendations from other experts. The latter scenario is advantageous for inexperienced users and can be used as a training method within the framework of application support.

What has been described above with reference to the computer program also applies analogously to the data processing device and vice versa.

Furthermore, the disclosure relates to a computer-implemented method for training a model based on artificial intelligence, the method comprising training the model such that, after training, the model is configured to determine information characterizing a treatment plan for the refractive eye surgery based on information characterizing a patient and information characterizing a refractive error of an eye of the patient that is to be treated by means of refractive eye surgery. A computer-implemented method can be understood as a method in which one, several or all steps of the method are carried out at least partially by a computer or a data processing device.

It is conceivable that the training process is structured in two stages. This means that in a first stage, the model can be pre-trained with predefined training data. In a second stage of the process, the model(s) can be fine-tuned or fine-adjusted. It is conceivable that the model(s) can be updated or further trained during ongoing operation (e.g., after each operation and/or after each recorded batch of surgical inputs). It is conceivable that the model(s) are trained in the second stage exclusively for those samples or input data for which the model predictions have a high uncertainty (i.e., an uncertainty exceeding a predetermined threshold). Additionally or alternatively, the model(s) can be trained in the second stage for those samples or input data that exhibit a distribution shift compared to previous input data (a new data representation leads to a covariance shift or other variations in the domain representation).

What has been described above with reference to the computer program and the data processing device also applies analogously to the computer-implemented method and vice versa.

Furthermore, the disclosure relates to a computer-readable medium comprising a model based on artificial intelligence, wherein the model is designed to, when the model is executed by a computer, cause the computer to use the model to, based on information characterizing a patient and information characterizing a refractive error of an eye of the patient that is to be treated by means of refractive eye surgery, to determine information characterizing a treatment plan for refractive eye surgery.

This means that a computer-readable medium can be provided that contains a computer program as defined above. The computer-readable medium can be any digital data storage device, such as a USB flash drive, a hard disk, a CD-ROM, an SD card, or an SSD card (or SSD drive/SSD hard disk).

The computer program does not necessarily have to be stored on such a computer-readable storage medium in order to be made available to the computer, but can also be obtained via the Internet or otherwise externally.

The computer-readable medium can therefore be a data signal comprising an artificial intelligence-based model for patient-specific planning of a refractive eye operation for a patient's eye by means of a laser, wherein the model is designed so that, when the model is executed by a computer, the computer uses the model to cause the computer to determine a setting of the laser characterizing the refractive eye operation based on information characterizing the patient and information characterizing a refractive error of the patient's eye that is to be corrected by means of the refractive eye operation.

What has been described above with reference to the computer program, the data processing device and the computer-implemented (training) method also applies analogously to the computer-readable medium and vice versa.

The above description can be summarized in other words and in a possible more concrete embodiment of the disclosure as described below, wherein the following description is not to be interpreted as limiting the disclosure. In summary, it can be stated that the provision of an AI-based assistance function (short: assistant) for the planning of refractive eye surgeries, optionally laser treatments, is proposed. The assistant can learn the relationship between diagnosis/patient data and the resulting treatment data through training with data from refractive surgery experts and/or the respective device user.

The assistant trained in this way can be made available for planning refractive eye surgeries and can suggest one or more treatment parameters to a user based on the patient's entered diagnostic data.

If the user changes a value suggested by the assistant during planning, it is conceivable that these changed data will be used for further training of the assistant.

It is also conceivable that the user can choose between several assistants during the planning process. These available assistants can be trained with user data and/or expert data.

The assistant can therefore help the user to complete their own planning more quickly by training the assistant with the user's data, and/or the assistant can support the user as a consultant based on recommendations from other physicians/experts. The latter scenario can be particularly helpful for inexperienced physicians and can be used as a training method for application support.

The disclosure therefore makes it possible to provide a comprehensive assistance function for the planning of refractive eye surgeries. It is conceivable that during the model training, the user subsequently evaluates the treatment parameters determined by the model. This assessment can be used to weight the training data.

It is conceivable that during model training, the user may be given the opportunity to point out or comment on certain dependencies between parameters. This information can be displayed to the user when using the wizard.

Connecting the devices and their planning software to a global IT data infrastructure (cloud) can facilitate the training and deployment of assistants between multiple users.

A training hierarchy can be used to train assistants. Assistants can be trained first with expert data and then fine-tuned with data from the respective user of the pre-trained model.

It is conceivable that the assistants learn and apply dependencies between treatment parameters. If certain dependencies are violated, the user can be notified.

It is conceivable that results from (possibly standardized) questionnaires completed by the patient are used by the assistant to suggest treatment parameters or a therapy modality.

It is conceivable that the results obtained via the questionnaire, initially presented as subjective data, are automatically objectified. These results can, for example, be converted into parameters and/or (optionally, K1) models can be used to convert text obtained via the questionnaire into embeddings that can be entered as additional inputs into one or more of the models described herein. A K1 model (e.g., RNNs, TCNs, transformers, etc.) can be used for processing Natural language can be used to convert the results obtained from the questionnaire, optionally presented as text, into more abstract information.

It is conceivable to offer the user the option to select for one or more treatment parameters whether the user would like to receive data from the assistant for this treatment parameter in future planning.

It is conceivable that the disclosed solution could be used as a training tool in application training courses. Based on the assistant's suggestions, the trainee can be taught how to select parameters or test their knowledge.

It is conceivable that the disclosed solution could be used in the context of device maintenance. There, the assistant could, for example, support the service technician in selecting parameters for the device configuration.

It is conceivable that the assistant supports the user in selecting the treatment modality (SMILE, LASIK, PRK, IOL, ICL, etc.). It is also conceivable that the assistant provides the user with a plan recommendation based on a predetermined rule. A rule could consist of only providing the assistant user with treatment recommendations generated by a model trained with the user's own treatment plans and/or expert treatment plans and/or treatment plans from colleagues in the user's country.

It is conceivable that the treatment plans issued by the assistant are commented on or evaluated by experts and/or automatically based on the achieved refractive patient outcome. The patient outcome can be determined by measuring the subjective and/or objective refraction of the eye. after surgery and/or by assessing patient satisfaction after refractive surgery, e.g. using questionnaires.

Furthermore, the disclosure relates to a computer-implemented method comprising determining, by means of an artificial intelligence-based model, information characterizing a treatment plan for refractive eye surgery based on information characterizing a patient and information characterizing a visual impairment of an eye of the patient to be treated by refractive eye surgery. The method comprises identifying at least one further patient from a plurality of patients and determining, by means of a recommendation system, further information characterizing a treatment plan for refractive eye surgery performed on the at least one identified patient based on the information characterizing the patient and/or the information characterizing the visual impairment of the patient to be treated by refractive eye surgery. The above description applies analogously to this method and vice versa.

An optional embodiment of the disclosure is described below with reference to Figures 1 to 3.

Fig. 1 shows a data processing system according to the disclosure, which is designed to execute a computer program according to the disclosure,

Fig. 2 shows a flowchart of a computer-implemented method according to the disclosure, which is implemented as the computer program according to the disclosure in the data processing system according to the disclosure shown in Fig. 1, and Fig. 3 shows a flowchart of another computer-implemented method according to the disclosure for training an artificial intelligence-based model, which is part of the computer program according to the disclosure.

The data processing system 1 shown in Figure 1, which can also be referred to as a data processing device, has a local computing unit 2 designed as a personal computer (PC) and a server 3 which is arranged remotely (i.e. spatially separated) from the PC 2 and is connected wirelessly (e.g. via the Internet) to the PC 2 for bidirectional data exchange.

For this purpose, the PC 2 comprises an input interface, comprising an input interface 6 for receiving user input (e.g., as a touch input and/or by means of a PC mouse and/or PC keyboard) and a data reception interface 7 for (in this case wirelessly) receiving data or information from the server 3. The PC 2 comprises an output interface, comprising an output interface 4 for (in this case visually) outputting information to a user of the PC 2 (which is designed as a display in this case) and a data output interface 8 for (in this case wirelessly) outputting data to or to the server 3.

The server 3 comprises a data reception interface 9 for (in this case wirelessly) receiving data or information from the PC 2. The server 3 comprises a data output interface 10 for (in this case wirelessly) outputting data to or from the PC 2.

The server 3 is designed to execute a part of a computer program or software by means of which a model 5 based on artificial intelligence and a recommendation system 11 are implemented. The PC 2 is designed to execute a further part of the computer program that is complementary to the part of the computer program executed by the server 3. which realizes user interaction. The following describes a sequence of a computer-implemented method that is realized by this computer program and executed by the data processing system 1 described above.

In a first step S1 of the method, a user input concerning information identifying a patient and information identifying a refractive error of an eye of the patient that is to be treated by means of refractive eye surgery is received via the input interface 6 of the PC 2.

The patient's identifying information includes information about the biomechanical stability and biometric characteristics of the treated eye, any pre-existing conditions or treatments of the treated eye, as well as the use and preference of visual aids, biological age, the patient's gender, pre-existing conditions, medications, lifestyle information, origin, and ethnicity. The information identifying the refractive error of the patient's treated eye, which is to be treated with refractive eye surgery, includes information about the manifest refraction and/or pachymetry of the treated eye.

In a second step S2 of the method, the information identifying the patient and the information identifying the refractive error of the patient's eye to be treated by refractive eye surgery are wirelessly output as a data signal via the data output interface 8 of the PC 2 to the data reception interface 9 of the server 3.

In a third step S3 of the method, the server 3 inputs the information characterising the patient received as a data signal and the received information characterising the refractive error of the patient's eye to be treated by means of the refractive eye surgery into the artificial intelligence-based model 5. The artificial intelligence-based Model 5 is designed for patient-specific planning of refractive eye surgery. For this purpose, Model 5 is trained to determine information characterizing a treatment plan for refractive eye surgery based on the information characterizing the patient and the information characterizing the refractive error of the patient's eye that is to be treated with refractive eye surgery.

Consequently, in a fourth step S4 of the method, the information characterizing the treatment plan for the refractive eye surgery is determined by means of the model 5 based on the information entered into the model 5 in the third step S3. The information characterizing the treatment plan for the refractive eye surgery comprises (at least) one treatment method from a list of predetermined treatment methods. The list from which the treatment procedure is selected by the Model 5 may include at least two of the following treatment procedures: femtosecond lenticule extraction (SMILE), photorefractive keratectomy (PRK), laser epithelial keratomileusis (LASEK), epithelial laser in situ keratomileusis (EpiLASIK), laser in situ keratomileusis (LASIK), and/or provision of an implant, optionally an intraocular lens (IOL) and/or an implantable contact lens (ICL), and/or corneal intrastromal implantation surgery (CISIS). In addition to the treatment procedure, the Model 5 may determine a target and/or a laser setting for performing the treatment procedure, optionally including information about a laser set, an optical zone, a laser energy, and/or a laser spot parameter. The specific target and/or setting of the laser may also be part of the information characterizing the treatment plan for refractive eye surgery.

In addition, in the fourth step S4, further information characterizing a treatment plan is determined. This information characterizes a refractive eye surgery that is to be performed on at least one other patient. was carried out. This can be the same type of information as the information determined using the Kl model, so that what has been described above with reference to it also applies analogously to the further information. To do this, the recommendation system first identifies at least one further patient from a large number of patients based on the information characterizing the patient and/or the information characterizing the refractive error of the patient's eye that is to be treated by means of refractive eye surgery. The recommendation system then determines the further information assigned to this identified patient. When identifying the further patient, the further patient from the large number of patients can be identified for whom a treatment result achieved with the refractive eye surgery performed on this further patient meets a predetermined (e.g. quality) criterion. It is conceivable that, additionally or alternatively, a similarity value is determined for each of the further patients. This can be done based on the information identifying the patient and/or the information identifying the refractive error of the patient's eye that is to be treated by refractive eye surgery, as well as information identifying the other patients, respectively, and/or information identifying a refractive error of one of the other patients' eyes that was treated by the refractive eye surgery performed on these patients. Thus, one or more of the other patients can be identified whose similarity value and/or whose treatment outcome fulfills/fulfills a predetermined criterion, e.g., whose similarity value is greater than a threshold value and/or whose treatment outcome lies within a tolerance range of a predetermined size around a to be achieved treatment outcome.

The recommendation system 11 can also be designed to determine a weighting factor. The recommendation system can be designed to use the weighting factor to determine intermediate results of the recommendation system and/or to rank predetermined therapeutic methods, predetermined treatment procedures and/or predetermined therapy-relevant information with which a therapy result achieved with the refractive eye surgery performed on this additional patient is likely to meet a predetermined criterion.

In a sub-step S41 of the fourth step, the result of the Kl model 5 can then be checked for plausibility using a plausibility check module 12. To this end, the plausibility check module 12 checks the plausibility of the determined information based on the determined further information. In concrete terms, this can mean, for example, that the plausibility check module 12 compares the information determined using the Kl model 5 with the further determined information and determines whether they at least partially match. In other words, it can be checked whether the (part of the) treatment plan determined using the Kl model 5 was at least similar or identical for similar or identified patients and/or whether a desired treatment outcome could be achieved for these identified patients.

In a fifth step S5 of the method, the specific information characterizing the treatment plan for the refractive eye surgery, the further specific information, the information (see above) concerning the identified further patients and/or a result of the plausibility check module 12 are again output in the form of a data signal via the data output interface 10 of the server 3 to the data reception interface 7 of the PC 2.

In a sixth step S6, based on the received information characterizing the treatment plan for the refractive eye surgery, the PC 2 controls its output interface 4 such that the output interface 4 displays the information characterizing the treatment plan for the refractive eye surgery. Furthermore, the further determined information, the information (see above) concerning the identified additional patients, and/or the result of the plausibility check module 12 can be displayed. Furthermore, the output interface can display the information determined by means of the CI model 5 in relation to a feasible parameter range. display, provide information on refractive planning quality, and/or issue a recommendation for refractive outcome optimization. All of this supports the system user in verifying the plausibility of the treatment plan determined by the computer program or the information determined by Kl-Model 5 and, if necessary, adopting it (see step S7).

In a seventh step S7 of the method, the user can again provide feedback via the input interface 6 of the PC 2 regarding the information characterizing the treatment plan for the refractive eye surgery (as a whole and/or regarding one or more of the above-described information contained in the information characterizing the treatment plan for the refractive eye surgery). Thus, the information characterizing the treatment plan for the refractive eye surgery can be accepted by the user directly or unchanged, or they can first modify it and then accept or release the modified information characterizing the treatment plan for the refractive eye surgery.

In an eighth step S8 of the method, the PC 2 generates a control signal based on the accepted information characterizing the treatment plan for the refractive eye surgery and outputs it to the laser (not shown). The control signal is designed to condition the laser to perform the refractive eye surgery in compliance with the information characterizing the treatment plan for the refractive eye surgery, as approved by the user.

In a ninth step S9 of the method, the user feedback (see step S7) and/or the changed and released information characterizing the treatment plan for the refractive eye surgery is output to the server 3 in the manner described above.

The following describes a training procedure for the artificial intelligence-based model 5 described above, the flow chart of which is shown in Figure 3 is shown. After performing the training procedure, model 5 is configured to execute the procedure described above.

In a first step S11 of the training method, a training data set is provided, which comprises a plurality of training examples. Each of the training examples, in turn, comprises information characterizing a patient, information characterizing a refractive error of one of this patient's eyes that is to be treated by refractive eye surgery, and information characterizing a treatment plan for the refractive eye surgery.

In a second step S12 of the training method, which is carried out for each of the training examples, the information characterizing the patient and the information characterizing the refractive error of the eye of this patient, which is to be treated by means of refractive eye surgery, of the respective training example are input into the model 5 and, based thereon, information characterizing a treatment plan for the refractive eye surgery is determined by means of the model 5.

In a third step S13 of the training process, which is again carried out for each of the training examples, the determined information characterizing the treatment plan for the refractive eye surgery is compared with the information characterizing the treatment plan for the refractive eye surgery stored in the respective training example. Based on the result of the comparison, parameters of model 5 are adjusted so that the information characterizing the treatment plan for the refractive eye surgery determined in the second step S2 increasingly approximates the information characterizing the treatment plan for the refractive eye surgery in the respective training example over the course of the training process. Thus, supervised (machine) learning is proposed. In a fourth step S14 of the training method, the model 5 trained as described above is made available to a user in the manner described above with reference to Figures 1 and 2.

In a fifth step S15 of the training method, which is executed, for example, by the server 3, the parameters of the trained model 5 are (further) adjusted based on the user feedback and/or the modified and released information characterizing the treatment plan for the refractive eye surgery, which is output to the server 3 in the ninth step S9 of the method described above with reference to Figure 2. In this way, the model 5 can be continuously improved.

A possible further development of the data processing system 1 described above with reference to Figure 1 and of the method described with reference to Figure 2 is explained below.

It is conceivable that several artificial intelligence-based models 5 are trained or taught using the training method described above. For this purpose, several training data sets can be provided, each containing various training examples. The training data sets can include training examples based on refractive eye surgeries performed in the past by various users. The training method described above, in particular the first three or four steps S1-S4, can be performed for each of the training data sets.

The result is a server 3 that includes several different computer-implemented models 5 based on artificial intelligence.

The method described with reference to Figure 2 can then comprise an initial step S0 in which the user selects one or more of the models 5 via the input interface 6 of the PC 2, which are used for the subsequent steps S1 - S9 for determining the treatment plan for the refractive eye surgery identifying information should be used. The models 5 can be part of a single or different computer programs.

List of reference symbols

1 Device or system for data processing

2 local processing unit, optional PC 3 remote processing unit, optional server

4 Output interface of the local processing unit

5 model based on artificial intelligence

6 Input interface of the local processing unit

7 Data reception interface of the local processing unit 8 Data output interface of the local processing unit

9 Data reception interface of the remote computing unit

10 Data output interface of the remote computing unit

11 Recommendation system

12 Plausibility module

SO - S9, S11 - S15 Process steps

Claims

Patent claims 1. Computer program comprising: - a model (5) based on artificial intelligence, wherein the computer program is designed to, when the computer program is executed by a computer (1), cause the computer (1) to do so by means of the model (5) based on: - information identifying a patient, and - information characterising a visual impairment of one of the patient's eyes to be treated by refractive eye surgery, information characterising a treatment plan for the refractive eye surgery, characterised in that the computer program comprises: - a recommendation system (11) which is designed to prompt the computer (1) to execute the computer program by means of the recommendation system (11) based on: - the information identifying the patient, and/or - the information characterising the refractive error of the patient's eye to be treated by means of the refractive eye surgery, to identify at least one further patient from a plurality of patients and to determine further information characterising a treatment plan for a refractive eye surgery performed on the at least one identified patient. 2. Computer program according to claim 1, wherein the computer program comprises a plausibility module (12) which is designed to, when the plausibility module (12) is executed by the computer (1), cause the computer (1) to execute a To check the plausibility of the specific information based on the specific further information. 3. Computer program according to claim 1 or 2, wherein the recommendation system (11) is designed to cause the computer (1) to execute the computer program by means of the recommendation system (11) to identify that further patient from the plurality of patients for whom a therapy result achieved with the refractive eye surgery performed on this further patient fulfills a predetermined criterion. 4. Computer program according to one of claims 1 to 3, wherein the recommendation system (11) is designed to cause the computer (1) to execute the computer program by means of the recommendation system (11): - a similarity value based on - the information identifying the patient, and/or - the information characterising the refractive error of the patient's eye to be treated by refractive eye surgery, and - information identifying the other patients, and/or - information identifying a refractive error in one eye of the other patients that was treated by the refractive eye surgery performed on these patients, and - to identify at least one other patient based on the determined similarity value. 5. Computer program according to one of claims 1 to 4, wherein the recommendation system (11) is designed to cause the computer (1) to execute the computer program by means of the recommendation system (11): - to determine a weighting factor, and - to use the weighting factor to determine intermediate results of the recommendation system (11) and/or to rank predetermined therapeutic methods, predetermined treatment procedures and/or predetermined therapy-relevant information with which a therapy result achieved with the refractive eye surgery performed on this further patient is likely to meet a predetermined criterion. 6. Computer program according to one of claims 1 to 5, wherein the information identifying the patient and, as far as related to claim 4, the information identifying the other patients, respectively, comprises information about: - biomechanical stability of the eye, - a biomechanical characteristic of the eye, - a previous eye disease, - pre-treatment of the eye, - a biological age, - one gender, - a pre-existing condition, - medication taken in the past and/or present, and/or - a lifestyle, origin and/or ethnicity of the respective patient. 7. Computer program according to one of claims 1 to 6, wherein the information characterizing the refractive error of the patient’s eye which is to be treated by means of the refractive eye surgery, and, as far as related to claim 4, the information which characterises the refractive error of the eye of the other patients which was treated by means of the refractive eye surgery performed on these patients, information about: - a manifest refraction, - a higher order aberration present before the refractive eye surgery, and/or - includes pachymetry. 8. A computer program according to any one of claims 1 to 7, wherein the information identifying the treatment plan for the refractive eye surgery and the further information identifying the treatment plan for the refractive eye surgery performed on the at least one identified patient each comprise: - a treatment procedure from a list of predetermined treatment procedures, the list optionally comprising at least two of the following treatment procedures: femtosecond lenticule extraction (SMILE), photorefractive keratectomy (PRK), laser epithelial keratomileusis (LASEK), epithelial laser in situ keratomileusis (EpiLASIK), laser in situ keratomileusis (LASIK), and/or providing an implant, optionally an intraocular lens (IOL) and/or an implantable contact lens (ICL), and/or corneal intrastromal implantation surgery (CISIS), and - optionally a target and/or a setting of the laser for carrying out the treatment procedure, optionally comprising information about a laser set, an optical zone, a laser energy and/or a laser spot parameter. 9. Device for data processing (1), wherein the device for data processing (1) is designed to use an artificial intelligence-based model (5) based on: - information identifying a patient, and - information characterising a visual impairment of one of the patient's eyes which is to be treated by means of refractive eye surgery, information characterising a treatment plan for the refractive eye surgery, characterized in that the data processing device (1 ) is designed to determine, by means of a recommendation system (11 ), based on: - the information identifying the patient, and/or - the information characterising the refractive error of the patient's eye to be treated by means of the refractive eye surgery, to identify at least one further patient from a plurality of patients and to determine further information characterising a treatment plan for a refractive eye surgery performed on the at least one identified patient. 10. Device for data processing (1) according to claim 9, wherein the device for data processing (1) is designed to check a plausibility of the determined information based on the determined further information by means of a plausibility check module (12). 11. A data processing device (1) according to claim 9 or 10, wherein the recommendation system (11) is configured to identify that further patient from the plurality of patients for whom a therapy result achieved with the refractive eye surgery performed on this further patient fulfills a predetermined criterion. 12. Device for data processing (1) according to one of claims 9 to 11, wherein the recommendation system (11) is designed to: - a similarity value based on - the information identifying the patient, and/or - the information characterising the refractive error of the patient's eye to be treated by refractive eye surgery, and - information identifying the other patients, and/or - information identifying a refractive error in one eye of the other patients that was treated by the refractive eye surgery performed on these patients, and - to identify at least one other patient based on the determined similarity value. 13. Device for data processing (1) according to one of claims 9 to 12, wherein the recommendation system (11) is designed to: - to determine a weighting factor, and - to use the weighting factor to determine intermediate results of the recommendation system (11) and/or to rank predetermined therapeutic methods, predetermined treatment procedures and/or predetermined therapy-relevant information with which a therapy result achieved with the refractive eye surgery performed on this further patient is likely to meet a predetermined criterion. 14. Device for data processing (1) according to one of claims 9 to 13, wherein the device for data processing (1) has an input interface for receiving the information characterizing the patient and/or information identifying a refractive error of the patient's eye that is to be treated by refractive eye surgery. 15. The data processing device (1) according to claim 14, wherein the input interface comprises an input interface (6), optionally comprising a display device, which is designed to receive a user input concerning the information characterizing the patient and/or the information characterizing a refractive error of the patient's eye that is to be treated by means of the refractive eye surgery. 16. Device for data processing (1) according to claim 14 or 15, wherein the input interface comprises a data receiving interface (7, 10) which is designed to receive a data signal comprising the information characterizing the patient and/or the information characterizing a refractive error of the patient's eye which is to be treated by means of the refractive eye surgery. 17. The data processing device (1) according to any one of claims 9 to 16, wherein the data processing device (1) comprises an output interface for outputting the determined information characterizing the treatment plan for the refractive eye surgery. 18. The data processing device (1) according to claim 17, wherein the output interface comprises an output interface (4) for outputting the determined information characterizing the treatment plan for the refractive eye surgery to a user of the data processing device (1). 19. Device for data processing (1) according to claim 18, wherein the output interface (4) is designed to - output feasible parameter ranges in relation to the information determined by the Kl model, - to provide information on the refractive planning quality, and/or - to issue a recommendation for refractive result optimization. 20. The data processing device (1) according to any one of claims 17 to 19, wherein the output interface comprises a data output interface (10) configured to output a data signal comprising the determined information characterizing the treatment plan for the refractive eye surgery. 21. A data processing device (1) according to any one of claims 17 to 20, wherein the device (1) is configured to determine a control signal for the laser based on the determined information characterizing the treatment plan for the refractive eye surgery and to output the determined control signal to the laser via the output interface for controlling the laser. 22. Device for data processing (1) according to one of claims 9 to 21, wherein the device (1) is designed to access a plurality of different computer-implemented models (5) based on artificial intelligence, by means of which the device for data processing (1) is each designed to be based on: - the information identifying the patient, and - the information characterising the refractive error of the patient's eye to be treated by refractive eye surgery, the information characterising the treatment plan for the refractive eye surgery. 23. Device for data processing (1) according to claim 22, insofar as it refers back to one of claims 14 to 16, wherein one or more of the models (5) is or are selectable via the input interface, optionally the input interface (6), of the device (1). 24. A computer-readable medium comprising a computer program according to any one of claims 1 to 8. 25. A computer-implemented method comprising: - Determining information characterizing a treatment plan for refractive eye surgery based on: - information identifying a patient, and - information characterising a refractive error of one of the patient's eyes to be treated by refractive eye surgery, by means of an artificial intelligence-based model (5), characterised in that the method comprises: - Identifying at least one further patient from a plurality of patients and determining further information characterizing a treatment plan for a refractive eye surgery performed on the at least one identified patient by means of a recommendation system (11) based on: - the information identifying the patient, and/or - the information that characterizes the patient's refractive error that is to be treated by refractive eye surgery.