US20220351410A1

US20220351410A1 - Computer assisted identification of appropriate anatomical structure for medical device placement during a surgical procedure

Info

Publication number: US20220351410A1
Application number: US17/681,963
Authority: US
Inventors: Krzysztof B. Siemionow; Cristian J. Luciano; Dominik Gawel; Marek KRAFT; Michal Trzmiel; Michal Fularz; Edwing Isaac MEJIA OROZCO
Original assignee: Holo Surgical Inc
Current assignee: Augmedics Inc
Priority date: 2018-08-10
Filing date: 2022-02-28
Publication date: 2022-11-03
Also published as: US11263772B2; EP3608870A1; US20200051274A1

Abstract

A method for computer assisted identification of appropriate anatomical structure for placement of a medical device, comprising: receiving a 3D scan volume comprising set of medical scan images of a region of an anatomical structure where the medical device is to be placed; automatically processing the set of medical scan images to perform automatic segmentation of the anatomical structure; automatically determining a subsection of the 3D scan volume as a 3D ROI by combining the raw medical scan images and the obtained segmentation data; automatically processing the ROI to determine the preferred 3D position and orientation of the medical device to be placed with respect to the anatomical structure by identifying landmarks within the anatomical structure with a pre-trained prediction neural network; automatically determining the preferred 3D position and orientation of the medical device to be placed with respect to the 3D scan volume of the anatomical structure.

Description

CROSS REFERENCE TO RELATED APPLICATIONS

This application is a continuation of U.S. patent application Ser. No. 16/537,645, filed Aug. 12, 2019, entitled “Computer Assisted Identification of Appropriate Anatomical Structure for Medical Device Placement During a Surgical Procedure,” the disclosure of which is hereby incorporated by reference.
U.S. patent application Ser. No. 16/537,645 claims priority to and the benefit of EP Patent Application No. 18188557.5, filed Aug. 10, 2018, entitled “Computer Assisted Identification of Appropriate Anatomical Structure for Medical Device Placement During a Surgical Procedure.”

TECHNICAL FIELD

The invention relates to computer assisted surgical navigation systems, in particular to a system and method for identifying appropriate anatomical structure for placement of a medical device, such as instrumentation or implant, during a surgical procedure, in particular related to neurological and general surgery procedures.

BACKGROUND

Image guided or computer assisted surgery is a surgical procedure where the surgeon uses trackable surgical instruments, combined with preoperative or intraoperative images (e.g., from computed tomography (CT) scanners), in order to provide the surgeon with surgical guidance during the procedure.

SUMMARY OF THE INVENTION

One of the disadvantages of known methods of image guided or computer assisted surgery is that they are not fully automatic. They require a specialized person to analyze the X-Ray, CT or NMR data and select a starting point for the procedure. Moreover they do not mention anything about the intraoperative CT allowing proper positioning during the surgery. In contrast, the invention, in certain embodiments, allows for fully automatic positioning and size determination in the 3D domain of the ongoing surgery thanks to usage of an intraoperative scanner and Artificial-Intelligence-based methods.
One aspect of the invention is a method for computer assisted identification of appropriate anatomical structure for placement of a medical device, comprising: receiving a 3D scan volume comprising set of medical scan images of a region of an anatomical structure where the medical device is to be placed; automatically processing the set of medical scan images to perform automatic segmentation of the anatomical structure; automatically determining a subsection of the 3D scan volume as a 3D region of interest by combining the raw medical scan images and the obtained segmentation data; automatically processing the ROI to determine the preferred 3D position and orientation of the medical device to be placed with respect to the anatomical structure by identifying landmarks within the anatomical structure with a pre-trained prediction neural network; automatically determining the preferred 3D position and orientation of the medical device to be placed with respect to the 3D scan volume of the anatomical structure.
The method may further comprise automatically identifying and storing the 3D position and orientation of the medical device placed by the surgeon in the anatomical structure during the surgical procedure, and using this information for further training of the prediction neural network in order to improve accuracy of the prediction neural network to subsequently identify the preferred positions and orientations to be suggested to the surgeon in successive surgical procedures.
The method may further comprise processing the scan images of the anatomical structures between the identified landmarks, and determining physical dimensions of the anatomical structures in the region of interest where the medical device is intended to be placed.
The method may further comprise determining preferred physical dimensions, the preferred physical dimensions including at least one of size, diameter and length, of the medical device to be placed depending on analyzed dimensions of the anatomical structure.
The received medical scan images may be collected from an intraoperative scanner.
The received medical scan images may be collected from a presurgical stationary scanner.
Another aspect of the invention is a computer-implemented system, comprising: at least one nontransitory processor-readable storage medium that stores at least one of processor-executable instructions or data; and at least one processor communicably coupled to at least one nontransitory processor-readable storage medium, wherein at least one processor is configured to perform the steps of the method as described herein.
These and other features, aspects and advantages of the invention will become better understood with reference to the following drawings, descriptions and claims.

BRIEF DESCRIPTION OF DRAWINGS

Various embodiments are herein described, by way of example only, with reference to the accompanying drawings, wherein:

FIG. 1 shows an overview of a training procedure in accordance with an embodiment of the invention;

FIG. 2A shows an image used in the system during the training procedures, in accordance with an embodiment of the invention;

FIG. 2B shows an image used in a system during the training procedures, in accordance with an embodiment of the invention;

FIG. 2C shows an image used in the system during the training procedures, in accordance with an embodiment of the invention;

FIG. 2D shows region of interest used in the process, in accordance with an embodiment of the invention;

FIG. 2E-1 shows 3 dimensional resizing of region of interest, in accordance with an embodiment of the invention;

FIG. 2E-2 shows 3 dimensional resizing of region of interest, in accordance with an embodiment of the invention;

FIG. 2F shows exemplary characteristic features localization, in accordance with an embodiment of the invention;

FIG. 2G shows exemplary results of artificial training database augmentation, in accordance with an embodiment of the invention;

FIG. 2H shows exemplary final implant localization, in accordance with an embodiment of the invention;

FIG. 3 shows an overview of a prediction procedure, in accordance with an embodiment of the invention;

FIG. 4 shows a prediction CNN architecture, in accordance with an embodiment of the invention;

FIG. 5 shows a flowchart of a training process for the prediction CNN, in accordance with an embodiment of the invention;

FIG. 6 shows a flowchart of an inference process for the prediction CNN, in accordance with an embodiment of the invention; and

FIG. 7 shows the structure of a computer system for implementing the method of FIG. 1, in accordance with an embodiment of the invention.

DETAILED DESCRIPTION OF THE INVENTION

The following detailed description is of the best currently contemplated modes of carrying out the invention. The description is not to be taken in a limiting sense, but is made merely for the purpose of illustrating the general principles of the invention.
The term “medical device” as used herein is understood to mean a surgical implant or an instrument, for example a catheter, instrument, a cannula, a needle, an anchor, a screw, a stent, a biomechanical device.
The invention is described below in detail with reference to an embodiment related to a neurological surgery, wherein a screw (as an example of a medical device) is placed, i.e. inserted, to a spine (as an example of an anatomical structure). A skilled person will realize that this embodiment can be extended to other applications as well, such as guidance for a medical device (e.g., instrumentation or implant) in other natural or artificial anatomical structures, for example blood vessels, biliary ducts, subthalamic nucleus, and components of solid organs like the heart (e.g., mitral valve), kidney (e.g., renal artery), and nerves (e.g., epidural space).
The automatic implant placement method as presented herein comprises two main procedures: a training procedure and a prediction procedure.
In certain embodiments, the training procedure, as presented in FIG. 1, comprises the following steps. First, in step 101, a set of DICOM (Digital Imaging and Communications in Medicine) images obtained with a preoperative or an intraoperative CT (Computed Tomography) or MRI (Magnetic Resonance Imaging) representing consecutive slices with visible tissues is received (such as one slice shown in FIG. 2A). Next, the received images are processed in step 102 to perform automatic segmentation of tissues, such as to determine separate areas corresponding to different tissues, such as vertebral body 16, pedicles 15, transverse processes 14 and/or spinous process 11, as shown in FIG. 2B. For example, this can be done by employing a method for segmentation of images disclosed in Applicant's European patent application EP17195826 filed Oct. 10, 2017 and published as EP 3 470 006 A1 on Apr. 17, 2019. Then, in step 103, the information obtained from DICOM images and the segmentation results is merged to obtain combined image comprising information about the tissue appearance and its classification (including assignment of structure parts to classes corresponding to different anatomy parts), for example in a form of a color-coded DICOM image, as shown in FIG. 2C. Alternatively, separate DICOM (FIG. 2A) and segmentation (FIG. 2B) images can be processed instead of the combined image. Next, in step 104, from the set of slice images a 3D region of interest (ROI) 18 is determined, that contains a volume of each pedicle 15 with a part of adjacent vertebral body 16 and surrounding tissues such as lamina 13, transverse process 14 and others, as shown in FIG. 2D. Then, in step 105, the 3D resizing of the determined ROI 18 is performed to achieve the same size of all ROI's stacked in the 3D matrices, each containing information about voxel distribution along X, Y and Z axes and the appearance and classification information data for each voxel, such as shown in FIG. 2E-1 or 2E-2. In other words, the voxels are small cuboidal volumes resembling points having 3D coordinates and a radiodensity value and classification assigned.
Next, in step 106, a training database is prepared manually, that comprises the previously determined ROIs and manually landmarked characteristic features corresponding to pedicle center 25 and screw tip 27 (or other anatomical structure and device points), such as shown in FIG. 2F.
Next, in step 107, the training database is augmented, for example with the use of a 3D generic geometrical transformation and resizing with random dense 3D grid deformations, as shown in FIG. 2G. Data augmentation is performed on the images to make the training set more diverse. The foregoing transformations are remapping the voxels positions in a 3D ROI 18 based on a randomly warped artificial grid assigned to the ROI 18 volume. A new set of voxel positions is calculated artificially warping the 3D tissue shape and appearance. Simultaneously, the information about the tissue classification is warped to match the new tissue shape and the manually determined landmarks positions 25, 27 are recalculated in the same manner. During the process, the value of each voxel, containing information about the tissue appearance, is recalculated in regards to its new position in ROI 18 with use of an interpolation algorithm (for example bicubic, polynomial, spline, nearest neighbor, or any other interpolation algorithm) over the 3D voxel neighborhood.
Next, in step 108, the obtained artificial database augmentation results are combined with the automatically recalculated landmarks, corresponding to the artificially augmented pedicle centers 25 and screw tips 27 (or other anatomical structure and device points), into a single database interpretable by a neural network.
Then, in step 109, the placement prediction model is trained with a neural network. In certain embodiments, a network with a plurality of layers is used, specifically a combination of convolutional and fully connected layers with ReLU activation functions or any other non-linear or linear activation functions. For example, a network such as shown in FIG. 4, according to a process such as shown in FIG. 5, can be used.
The training database may also comprise data from actually performed surgical procedures. The system may automatically identify and store the 3D position and orientation of the medical device actually inserted by the surgeon in the anatomical structure during the surgical procedure, for further training the prediction neural network (400) in order to improve its performance to subsequently identify the preferred positions and orientations. Therefore, the system may operate like a closed feedback loop.
In certain embodiments, the prediction procedure, as presented in FIG. 3, comprises the following steps. First, in step 301, a 3D scan volume is received, comprising a set of DICOM (Digital Imaging and Communications in Medicine) images of a region of the anatomical structure where the medical device is to be placed. The 3D scan volume can be obtained with a preoperative or an intraoperative CT (Computed Tomography) or Mill (Magnetic Resonance Imaging). The set of DICOMs representing consecutive slices of a spine is received (such as one slice shown in FIG. 2A). Next, the received images are processed in step 302 to perform automatic segmentation of tissues of the anatomical structure, such as to determine separate areas corresponding to different tissues, such as vertebral body 16, pedicles 15, transverse processes 14, lamina 13 and/or spinous process 11, as shown in FIG. 2B. For example, this can be done by employing a method for segmentation of images disclosed in Applicant's European patent application EP17195826 filed Oct. 10, 2017 and published as EP 3 470 006 A1 on Apr. 17, 2019, incorporated herein by reference in its entirety. Then, in step 303, the information obtained from DICOM images and the segmentation results is merged to obtain combined image comprising information about the tissue appearance and its classification, for example in a form of a color-coded DICOM image, as shown in FIG. 2C. Alternatively, separate DICOM (FIG. 2A) and segmentation (FIG. 2B) images can be processed instead of the combined image. Next, in step 304, from the 3D scan volume a 3D region of interest (ROI) 18 is automatically determined. For example, the ROI 18 may contain a volume of each pedicle 15 with a part of adjacent vertebral body and surrounding tissues, as shown in FIG. 2D. Then, in step 305, the 3D resizing of the determined ROI 18 is performed to achieve the same size of all ROI's stacked in the 3D matrices. Each 3D matrix contains information about voxel distribution along X, Y and Z axes with bone density and classification information data for each voxel, such as shown in FIG. 2E-1 or 2E-2. Therefore, steps 301-305 are performed in a way similar to steps 101-105 of the training procedure of FIG. 1.
Next, in step 306, the preferred placement is predicted automatically by processing the resized ROI to determine the preferred 3D position and orientation of the medical device to be placed with respect to the anatomical structure, by means of the pretrained prediction CNN 400, according to the prediction process presented in FIG. 6. The prediction CNN 400 is configured to identify landmarks within the anatomical structure, such as pedicle center 25 and screw tip 27.
Next, in step 307, the predicted screw tip 25 and pedicle center 27 (or other anatomical structure and device landmarks) positions within the ROI are backward recalculated to meet the original ROI size and positions from input DICOM dataset to recreate and ensure a correct placement in original volume.
In step 308 the information about the global coordinate system (ROI position in the DICOM dataset) and local ROI coordinate system (predicted screw tip and pedicle center positions inside the ROI) is recombined.
Then, in step 309, the preferred device positioning in the 3D space is calculated, based on two landmarks corresponding to pedicle center 25 and screw tip 27, as shown in FIG. 2F.
Anatomical knowledge and preferred device positioning allow for the calculation of a preferred device's physical dimensions, for example screw positioning in the vertebra. With the semantic/anatomical segmentation results and pedicle center 25 location available, in step 310, automated computation of device physical dimensions, such as the diameter, is possible. Proceeding in the coronal direction, forward and backward from the pedicle center landmark 25 along the pedicle, the slice for which the inscribed circle diameter will be the smallest can easily be found. A fraction of this diameter corresponds directly to the inserted device maximum allowed diameter with a necessary safety margin that can be easily defined by the user of the system.
Enabling selection of a specific element in the available series of types also requires determination of device physical dimensions such as the length. This too can be easily computed automatically using the device insertion trajectory information provided by the neural network. The line going through the estimated landmarks (bone anchor tip 27, pedicle center 25) represents the trajectory of the device, which can be expressed as a 3D path, in the case of 2 landmarks it will be line model. Given the trajectory of a medical device to be inserted and an anatomical structure being a target, the entry and exit points could be calculated using automated 3D image analysis. For example, given the 3D line model and a 3D shell of the shape of the anatomical part being a target of device insertion extracted using morphological gradient in 3D (a single voxel thick surface of all solids in the volume), the entry and exit points of the trajectory are located at the two shell voxels (XX, YY) that are closest to the line (trajectory T) at each end, for example such as shown in FIG. 2F.
Next, in step 311, the output is visualized, for example such as shown in FIG. 2H, including the device 31 to be inserted.
FIG. 4 shows a convolutional neural network (CNN) architecture 400, hereinafter called the prediction CNN, which is utilized in certain embodiments of the method of the invention for prediction of device placement. The network performs device localization task using at least one input as a 3D information about the appearance (radiodensity) and the classification for each voxel in a 3D ROI.
The left side of the network is a contracting path, which includes convolution layers 401 and pooling layers 402, and the right side is a regression path which includes fully connected layers 403 and the output layer 404.
One or more 3D ROI's can be presented to the input layer of the network to learn reasoning from the data.
The convolution layers 401 can be of a standard kind, the dilated kind, or a combination thereof, with ReLU, leaky ReLU or any other kind of activation function attached.
The fully connected layers 403 can have Linear, ReLU or any other kind of activation function attached.
The output layer 404 also denotes the fully connected layer with the loss function, for example the loss function can be implemented as mean squared error or another metric.
The architecture is general, in the sense that adopting it to ROI's of different size is possible by adjusting the size (resolution) of the layers. The number of layers and number of filters within a layer is also subject to change, depending on the requirements of the application, for example as presented in Applicant's European patent application EP17195826.
The final layer for the device placement defines the preferred device position and orientation along X, Y and Z axes in 3D ROI. Prediction is based on the model trained from the manually prepared examples during the training process, for example in case of screw insertion, preferred position of the pedicle center 25 and screw tip 27.
FIG. 5 shows a flowchart of a training process, in accordance with certain embodiments, which can be used to train the prediction CNN 400. The objective of the training for the prediction CNN 400 is to tune the parameters of the prediction CNN 400 such that the network is able to predict preferred guidance for the device.
The training database may be separated into a training set used to train the model, a validation set used to quantify the quality of the model, and a test set.
The training starts at 501. At 502, batches of training ROI's are read from the training set, one batch at a time.
At 503 the ROI's can be additionally augmented. Data augmentation is performed on these ROI's to make the training set more diverse. The input/output data is subjected to the combination of transformations from the following set: rotation, scaling, movement, horizontal flip, additive noise of Gaussian and/or Poisson distribution and Gaussian blur, volumetric grid deformation, etc. or could be augmented with the use of generative algorithm such as Generative Adversarial Networks for example.
At 504, the ROI's are then passed through the layers of the CNN in a standard forward pass. The forward pass returns the results, which are then used to calculate at 505 the value of the loss function—the difference between the desired and the computed outputs. The difference can be expressed using a similarity metric (e.g., mean squared error, mean average error or another metric).
At 506, weights are updated as per the specified optimizer and optimizer learning rate using Gradient Descent methods (e.g., Stochastic Gradient Descent, Adam, Nadam, Adagrad, Adadelta, RMSprop).
The loss is also back-propagated through the network, and the gradients are computed. Based on the gradient values, the network's weights are updated. The process (beginning with the ROI's batch read) is repeated continuously until the end of the training session is reached at 507.
Then, at 508, the performance metrics are calculated using a validation dataset—which is not explicitly used in training set. This is done in order to check at 509 whether or not the model has improved. If it is not the case, the early stop counter is incremented at 514 and it is checked at 515 if its value has reached a predefined number of epochs. If so, then the training process is complete at 516, since the model has not improved for many sessions now.
If the model has improved, the model is saved at 510 for further use and the early stop counter is reset at 511. As the final step in a session, learning rate scheduling can be applied. The sessions at which the rate is to be changed are predefined. Once one of the session numbers is reached at 512, the learning rate is set to one associated with this specific session number at 513.
Once the training is complete, the network can be used for inference (i.e., utilizing a trained model for prediction on new data).
FIG. 6 shows a flowchart of an inference process for the prediction CNN 400 in accordance with certain embodiments of the invention.
After inference is invoked at 601, a set of ROI's is loaded at 602 and the prediction CNN 400 and its weights are loaded at 603.
At 604, one batch of ROI's at a time is processed by the inference server.
At 605, the images can be preprocessed (e.g., normalized)
At 606, a forward pass through the prediction CNN 400 is computed.
At 607, a postprocess prediction is done.
At 608, if not all batches have been processed, a new batch is added to the processing pipeline until inference has been performed on all input ROI's.
Finally, at 609, the inference results are saved and can be recalculated to provide an output in a form of preferred device position.
The functionality described herein can be implemented in a computer system 700, such as shown in FIG. 7. The system 700 may include at least one nontransitory processor-readable storage medium 710 that stores at least one of processor-executable instructions 715 or data; and at least one processor 720 communicably coupled to the at least one nontransitory processor-readable storage medium 710. The at least one processor 720 may be configured to (by executing the instructions 715) perform the procedure of FIG. 1.
While the invention has been described with respect to a limited number of embodiments, it will be appreciated that many variations, modifications and other applications of the invention may be made. Therefore, the claimed invention as recited in the claims that follow is not limited to the embodiments described herein.

Claims

1.-7. (canceled)

8. A method, comprising:

receiving a set of three-dimensional (3D) scan volumes each including image data of an anatomical structure with a medical device implanted in the anatomical structure;

processing each of the set of 3D scan volumes to obtain segmentation data that identifies different anatomical parts of the anatomical structure of each of the set of 3D scan volumes;

merging, for each of the set of 3D scan volumes, the segmentation data of that 3D scan volume with the image data of that 3D scan volume to produce combined image data for that 3D scan volume;

identifying a 3D region of interest (ROI) in the combined image data of each of the set of 3D scan volumes that includes a subsection of the anatomical structure of that 3D scan volume with the medical device implanted therein;

generating a training database including (1) the 3D ROI identified for each of the set of 3D scan volumes and (2) marked characteristic features associated with a set of landmarks in the 3D ROI identified for each of the set of 3D scan volumes, the set of landmarks including an anatomical part of the anatomical structure and a portion of the medical device implanted in the anatomical structure; and

training, using the training database, a prediction neural network model to identify preferred positions and orientations for placement of the medical device in the anatomical structure.

9. The method of claim 8, further comprising:

augmenting the training database by: transforming the 3D ROI identified for each of the set of 3D scan volumes using one or more of: rotation, scaling, movement, horizontal flip, additive noise of Gaussian or Poisson distributions and Gaussian blur, volumetric grid deformation, or a generative algorithm, and generating an augmented training database including the 3D ROIs and the transformed 3D ROIs; and

training the prediction neural network model using the augmented training database.

10. The method of claim 9, wherein the augmenting the training database further includes recalculating a position of the set of landmarks in each of the transformed 3D ROIs and associating the recalculated position of the set of landmarks with the transformed 3D ROIs.

11. The method of claim 8, further comprising:

resizing, for each of the set of 3D scan volumes, the 3D ROI identified for that 3D scan volume such that each 3D ROI has the same size,

the training database including the resized 3D ROI for each of the set of 3D scan volumes and the marked characteristic features associated with the set of landmarks in the resized 3D ROI.

12. The method of claim 8, further comprising determining, using the trained prediction neural network model, a preferred placement of the medical device in an anatomical structure of a patient.

13. The method of claim 12, wherein the determining the preferred placement of the medical device in the anatomical structure of the patient includes:

determining, based on a 3D scan volume of anatomical structure of the patient and segmentation data of the anatomical structure of the patient, a 3D ROI in the 3D scan volume of the anatomical structure of the patient for placement of the medical device;

processing, using the trained prediction neural network model, the 3D ROI of the patient to identify the set of landmarks in the 3D ROI of the patient; and

determining a 3D position and orientation of the medical device for the preferred placement of the medical device based on the set of landmarks.

14. The method of claim 13, further comprising visualizing the 3D position and orientation of the medical device for the preferred placement of the medical device with respect to the anatomical structure.

15. The method of claim 13, further comprising

identifying, after the medical device has been placed in the anatomical structure of the patient, an actual 3D position and orientation of the medical device; and

further training the prediction neural network model based on the actual 3D position and orientation of the medical device.

16. The method of claim 8, wherein the anatomical part of the set of landmarks includes a pedicle, and the portion of the medical device of the set of landmarks includes a tip of the medical device.

17. A method, comprising:

identifying a 3D region of interest (ROI) in each of the set of 3D scan volumes that includes a subsection of the anatomical structure of that 3D scan volume with the medical device implanted therein;

generating a training database including (1) the 3D ROI identified for each of the set of 3D scan volumes and (2) marked characteristic features associated with a set of landmarks in the 3D ROI identified for each of the set of 3D scan volumes, the set of landmarks including an anatomical part of the anatomical structure and a portion of the medical device implanted in the anatomical structure;

training, using the training database, a prediction neural network model to identify preferred positions and orientations for placement of the medical device in the anatomical structure; and

determining, using the trained prediction neural network model, a preferred placement of the medical device in an anatomical structure of a patient.

18. The method of claim 17, further comprising:

19. The method of claim 18, wherein the augmenting the training database further includes recalculating a position of the set of landmarks in each of the transformed 3D ROIs and associating the recalculated position of the set of landmarks with the transformed 3D ROIs.

20. The method of claim 17, further comprising:

21. The method of claim 17, further comprising determining, using the trained prediction neural network model, a preferred placement of the medical device in an anatomical structure of a patient.

22. An apparatus, comprising:

a memory; and

a processor operatively coupled to the memory, the processor configured to:

receive a set of three-dimensional (3D) scan volumes each including image data of an anatomical structure with a medical device implanted in the anatomical structure;

process each of the set of 3D scan volumes to obtain segmentation data that identifies different anatomical parts of the anatomical structure of each of the set of 3D scan volumes;

merge, for each of the set of 3D scan volumes, the segmentation data of that 3D scan volume with the image data of that 3D scan volume to produce combined image data for that 3D scan volume;

identify a 3D region of interest (ROI) in the combined image data of each of the set of 3D scan volumes that includes a subsection of the anatomical structure of that 3D scan volume with the medical device implanted therein;

generate a training database including (1) the 3D ROI identified for each of the set of 3D scan volumes and (2) marked characteristic features associated with a set of landmarks in the 3D ROI identified for each of the set of 3D scan volumes, the set of landmarks including an anatomical part of the anatomical structure and a portion of the medical device implanted in the anatomical structure; and

train, using the training database, a prediction neural network model to identify preferred positions and orientations for placement of the medical device in the anatomical structure.

23. The apparatus of claim 22, wherein the processor is further configured to determine, using the trained prediction neural network model, a preferred placement of the medical device in an anatomical structure of a patient.

24. The apparatus of claim 22, wherein the processor is further configured to:

determine, based on a 3D scan volume of anatomical structure of the patient and segmentation data of the anatomical structure of the patient, a 3D ROI in the 3D scan volume of the anatomical structure of the patient for placement of the medical device;

process, using the trained prediction neural network model, the 3D ROI of the patient to identify the set of landmarks in the 3D ROI of the patient; and

determine a 3D position and orientation of the medical device for the preferred placement of the medical device based on the set of landmarks.

25. The apparatus of claim 24, further comprising a display device communicably coupled to the processor, wherein the display device is configured to display the 3D position and orientation of the medical device for the preferred placement of the medical device with respect to the anatomical structure.

26. The apparatus of claim 25, wherein the processor is further configured to:

identify, after the medical device has been placed in the anatomical structure of the patient, an actual 3D position and orientation of the medical device; and

further train the prediction neural network model based on the actual 3D position and orientation of the medical device.

27. The apparatus of claim 22, wherein the anatomical part of the set of landmarks includes a pedicle, and the portion of the medical device of the set of landmarks includes a tip of the medical device.