WO2018132090A1

WO2018132090A1 - Methods for processing three dimensional body images

Info

Publication number: WO2018132090A1
Application number: PCT/US2017/012891
Authority: WO
Inventors: Ignatius Dewet DIENER; Marcus Foster; David Greer; Kevin KERAUDREN; Brandon WHITCHER
Original assignee: Klarismo Inc
Current assignee: Klarismo Inc
Priority date: 2017-01-10
Filing date: 2017-01-10
Publication date: 2018-07-19
Anticipated expiration: 2019-07-10

Abstract

Within this disclosure are new methods of producing three-dimensional models of a body based on images. In one embodiment, two-dimensional MRI images are used to create a three-dimensional model of a body. In one embodiment, the three-dimensional model is segmented and landmarked to provide reference points for allowing for analysis. In some embodiments, volumetric values are calculated to provide significant details about the body. In one embodiment, the body is a human body.

Description

METHODS FOR PROCESSING THREE DIMENSIONAL BODY IMAGES

Technical Field

This disclosure relates to body imaging and analysis. In particular, this disclosure relates to processing MRI images for producing three-dimensional images.

Background

While advances in medicine have prolonged life by curing and preventing diseases, various types of cancer, genetic defects, etc. , one of the most important factors is early detection. Medical diagnosis is critical for health treatment and prevention. Detecting the early stages of diseases allows medical professionals to start treatment early to give a patient the best chance of recovery and survival. To that end, advances in medical diagnosis are just as important as medical treatment.

One key area of medical diagnosis is medical imaging. Some medical imaging reveals internal structures hidden by the skin and bones that escape surface level physical examinations. Medical imaging creates opportunities for establishing a database of normal anatomy and physiology for identifying potential abnormalities.

Common medical imaging techniques include X-ray, Magnetic Resonance Imaging (MRI), and Computerized Tomography (CT) scans. One advantage MRI scans have over CT scans is that MRI does not expose the patient to ionizing radiation that could cause potential side effects. MRI scans provide a high level of detail, resolution, and clarity of the anatomy and physiology of the human body. MRI imaging collects an incredible amount of detailed information about the entire body. MRI scans are traditionally done because there is a specific need to investigate, e.g., cancer concerns, diseases, organ issues, etc. MRI scans are not traditionally used for a routine examination. However, one drawback to MRI imaging is that most medical professionals focus on a small specific part of the body. At present, much of the information contained within an MRI image is left unused.

Most patients are not capable of using an MRI image for their own personal benefit outside of medical treatment. Patients will even sometimes receive a disk of their MRI images but not have the ability and tools to fully extrapolate the information they have about their body. An MRI scan can provide incredibly useful information about one's entire body and could be used to develop diets, exercise plans, treatments, etc. Another limitation of the current state of the art is that radiologists perform qualitative assessments of body images whereas analyzing the images quantitatively would provide for improvements in speed, efficiency, accuracy, etc.

There exists a need for medical imaging providing in-depth details about a body. There exists a need for medical imaging producing three-dimensional representations. There exists a need for medical imaging providing segmentation of the body. There exists a need for medical imaging providing volumetric measurements. There also exists a need for new ways to use MRI images and data. Brief Description of the Drawing

Fig. 1 is a diagram of a non-limiting example of how a spring like model is used for creating a face by assigning scores to reference points.

Detailed Description

Disclosed herein is a new method for producing, processing, creating, analyzing, and/or creating medical images. In one embodiment, the medical images are three-dimensional. In one embodiment, landmarks are detected. In one embodiment, a body is segmented. In one embodiment, volumetric values are calculated. Disclosed herein is a new three-dimensional image of a body. In one embodiment, the body is a human body. In one embodiment, the body is a dog. In one embodiment, the body is a horse. In one embodiment, the three-dimensional images of a body is segmented.

Disclosed herein is a new method of using MRI images. In one embodiment, a plurality of MRI scans are used to construct a three-dimensional model from two-dimensional scans. In one embodiment, a cloud database platform is used to process MRI images.

Disclosed herein is a new method of medical diagnosis. In one embodiment, the method disclosed herein comprises using images produced by the methods disclosed herein to diagnose a disease or condition. In one embodiment, a database of three-dimensional representations is created for identifying anomalies and/or abnormalities.

Disclosed herein is a method of measuring body volume, comprising:

Collecting a plurality of two-dimensional images for a body;

Processing the plurality of two-dimensional body images into a three-dimensional representation of the body; Assigning landmarks to points within the three-dimensional representation of the body;

Registering the three-dimensional representation of the body;

Segmenting the three-dimensional representation of the body into segments; and

Calculating the volume of at least one of the segments.

As used herein, the term "collecting" means gathering, receiving, transferring, downloading, and/or finding. In one embodiment, collecting comprises receiving information from a cloud database. In one embodiment, collecting comprises transferring information from a disk onto a computer. In one embodiment, collecting comprises transferring information from a USB drive onto a computer. In one embodiment, collecting is streaming.

In one embodiment, collecting comprises receiving images from an MRI machine. An RI machine uses magnetic energy causing the change of the orientation of protons in the body. As the protons return to their original orientation, they produce radio signals which are recorded. Protons in different tissues of the body behave differently producing different radio signals allowing an MRI machine to differentiate the different tissues of the body. The intensity of the received signal is then plotted on a grey scale and cross sectional images are built up. In one embodiment, the MRI machine stores images in DICOM format. Within the context of this disclosure, collecting encompasses passively receiving information in which information is collected without request. Within the context of this disclosure, collecting encompasses actively receiving information in which information is requested. As used herein, the term "plurality" refers to more than one. In one embodiment, there is a plurality of bodies. In one embodiment, there is a plurality of images for one body. In one embodiment, there is a plurality of two-dimensional images of the body. In one embodiment, there is a plurality of planar slices of the body. In one embodiment, there is a plurality of parts of the same body. In one embodiment, there is a plurality of images for a plurality of bodies. In one embodiment, there is a plurality of images for one segment of a body.

As used herein, the term "two-dimensional" refers to an object or image appearing to have width and length, i.e., area, within a plane. Area is the quantity expressing the extent of a two-dimensional figure or shape in a plane. In one embodiment, two-dimensional comprises cartesian coordinates. In one embodiment, an image is in the x and y axis. In one embodiment, a body is in the x and z axis. In one embodiment, a body is in the y and z axis. In one embodiment, two dimensional comprises polar coordinates.

As used herein, the term "image" refers to a likeness or representation of a mass made visible. In one embodiment, an image is a digital image. In one embodiment, an image is a picture of a human body. In one embodiment, an image is from an RI machine. In one embodiment, an image is a collection of pixels. In one embodiment, the image is in a JPEG format. In one embodiment, the image is in a DICOM format. In one embodiment, the image is in a NlfTI format. As used herein, the term "two-dimensional image" refers to a presentation of mass present within a particular plane, e.g., cross-section of a body. In one embodiment, the two- dimensional image comprises cartesian coordinates, e.g., x, y, and z axis. In one embodiment, the two-dimensional image is a picture. In one embodiment, the two- dimensional image is a digital image on a screen.

As used herein, the term "body" refers to the physical structure and material substance of a physical mass, e.g., an organism. In one embodiment, the body is living. In one embodiment, the body is not living. In one embodiment, the body is a human. In one embodiment, the body is a dog. In one embodiment, the body is a horse. In one embodiment, the body comprises organs. In one embodiment, the body comprises a skeleton. In one embodiment, the body comprises fat. In one embodiment, the body comprises muscles. In one embodiment, the body comprises subcutaneous fat. In one embodiment, the body comprises adipose tissue. In one embodiment, the body comprises visceral fat. In one embodiment, the body comprises iliopsoas muscles.

As used herein, the term "image of a body" refers to a representation of a physical structure and material substance of a physical mass, e.g., an organism. In one embodiment, the image of a body is an embodiment of an MRI scan. In one embodiment, the image of a body is an embodiment of a CT scan. In one embodiment, the image of a body is an internal image. In one embodiment, the image of a body is of a human body or a portion of a human body.

As used herein, the term "processing" refers to treating a thing through a series of steps. In one embodiment, processing comprises using a computer algorithm. In one embodiment, processing comprises producing an image from another image. In one embodiment, processing comprises using a processor. In one embodiment, processing comprises manually manipulating an image. In one embodiment, processing comprises creating three-dimensional images from a series of two-dimensional crosscut images. In one embodiment, processing comprises analyzing segments of a body.

As used herein, the term "three-dimensional" refers to an object or image appearing to have length, depth, and breadth, i.e., volume. Volume is the amount of space occupied by an object. In one embodiment, volume is measured in cubic meters (m³). In one embodiment, three dimensional comprises cartesian coordinates. In one embodiment, an image has multiple x, y, and z coordinates. In one embodiment, volume has polar coordinates. As used herein, the term "representation of the body" refers to a concrete portrayal of the physical structure and material substance of a physical mass, e.g., an organism. In one embodiment, the representation of the body is two-dimensional. In one embodiment, the representation of the body is three-dimensional. In one embodiment, the representation of the body is a representation of a human body. In one embodiment, the representation of the body is a picture. In one embodiment, the representation of the body is a digital image. In one embodiment, the representation of the body is a representation of the skeleton. In one embodiment, the representation of the body is an MRI scan. In one embodiment, the representation of the body is a CT scan. In one embodiment, the representation of the body is an X-ray.

As used herein, the term "assigning" refers to categorizing or labeling specific things, e.g., points, areas, or volumes the body. In one embodiment, assigning refers to marking areas of the body with a machine learning program. In one embodiment, assigning refers to detecting landmarks.

As used herein, the term "landmark" refers to a particular reference part of a thing. In one example, an area of the body where a bone is on the surface. Landmarks are often used as a reference point of the body to find other structures. In one embodiment, landmarks are used to measure proportion. In one embodiment, landmarks are used to find forms. In one embodiment, a landmark is a bone joint. In one embodiment, a landmark is the vertebrae. In one embodiment, a landmark is the armpit. In one embodiment, a landmark is a spine curve. In one embodiment, a landmark is a shoulder; left, right, or both. In one embodiment, a landmark is a hip; left, right, or both. In one embodiment, a landmark is a knee; left, right, or both. In one embodiment, a landmark is an ankle; left, right, or both. In one embodiment, the landmarks are based on a coordinate system unique to an individual body. As used herein, the term "assigning a landmark" refers to designating a particular reference part of a thing. In one embodiment, assigning a landmark comprises labeling a bone joint. In one embodiment, assigning a landmark comprises labeling the centerline of the front of the body. In one embodiment, assigning a landmark comprises labeling the centerline of the back. In one embodiment, assigning a landmark comprises labeling the shoulder; left, right, or both. In one embodiment, assigning a landmark comprises labeling the hip; left right, or both. In one embodiment, assigning a landmark comprises labeling the knee; left, right, or both. In one embodiment, assigning a landmark comprises labeling the ankle; left, right, or both. As used herein, the term "point within the three-dimensional representation of the body" refers to a specific position within a portrayal of a physical structure with reference to physical structure's length, width, and height/depth. In one embodiment, the point within the three-dimensional representation of the body is a landmark. In one embodiment, the point within the three-dimensional representation of the body is a bone joint. In one embodiment, the point within the three-dimensional representation of the body is a body part. In one embodiment, the point within the three-dimensional representation of the body is the chest.

As used herein, the term "registering" refers to matching and/or aligning two objects based on a set of features. In one embodiment, "registering" means aligning two images, such as landmarks of a representation of a body. In one example, registering means finding the warping transforming an image so that it best fits another image. In one embodiment, registering means finding a transformation mapping an image onto another image. In one embodiment, registering allows subsequent processes to recognize an image quicker and more efficiently. In one embodiment, registering comprises recording an image of a specific landmark common to multiple bodies. In one embodiment, registering comprises utilizing a cloud database platform. In one embodiment, registering comprises implementing non-rigid registration. In one embodiment, non-rigid registration is elastic deformation. As used herein, the term "segmenting" refers to dividing, cutting, isolating, and/or segregating a thing into separate pieces or sections. Segmenting allows one to focus on an area of interest and study that area in more detail. In one embodiment, segmenting comprises dividing pieces of a body. In one embodiment, segmenting comprises dividing images of a body. In one embodiment, segmenting comprises dividing a body into a left and right half. In one embodiment, segmenting comprises dividing the body into the head, torso, left leg, and right leg. In one embodiment, segmenting comprises dividing the torso into the chest, back, abdomen, and shoulders. In one embodiment, segmenting comprises dividing the leg muscles into the upper and lower segments. In one embodiment, segmenting comprises dividing the leg muscles into left and right segments. In one embodiment, segmenting comprises dividing the leg bones into the femur and tibia. In one embodiment, segmenting comprises using atlas based segmentation of the kidneys. In one embodiment, segmenting comprises using atlas based segmentation of the liver.

As used herein, the term "segment" refers to an individual section or piece of a thing. In one embodiment, the segment is attached to the thing. In one embodiment, the segment is separated from the thing. In one embodiment, a body is separated. In one embodiment, a segment of the body is an arm. In one embodiment, a segment of the body is the torso. In one embodiment, a segment of the body is separated into more segments. In one embodiment, an arm is segmented into fingers, hand, elbow, shoulder, etc. In one embodiment, a segment is tissue of the body. As used herein, the term "calculating" refers to determining a value through a computation or computations. In one embodiment, calculating comprises using a machine learning program. In one embodiment, calculating comprises manual computations. In one embodiment, calculating comprises determining the volume of a segment. In one embodiment, calculating comprises determining the amount of fat in a torso. In one embodiment, calculating comprises measuring the muscle mass of an arm.

As used herein, the term "machine learning program" refers to a type of artificial intelligence providing an apparatus, e.g., a computer, the ability to comprehend material without being explicitly instructed. In one embodiment, a machine learning program is taught to learn languages. In one embodiment, a machine learning program predicts the weather based on weather changes. In one embodiment, a machine learning program filters spam. In one embodiment, a machine learning program is taught to mark landmarks in different bodies. In one embodiment, the machine learning program is taught to detect centroids of the vertebrae (e.g., L5 - T8). In one embodiment, a machine learning program automates the methods disclosed herein. In one embodiment, the number of variables determines the number of features. In one embodiment, multivariables result in either a classification or regression analysis.

As used herein, the term "volume" refers to an amount of three-dimensional space. Examples of measurements of volume include, but are not limited to, cubic inches, cubic feet, cubic centimeters, cubic meters, cubic millimeters, and cubic liters. In one embodiment, volume correlates with the distribution of weight within the body. In one embodiment, volume comprises the fat content of the body. In one embodiment, volume comprises a measurement of total body fat. In one embodiment, volume comprises a measurement of total muscle tissue. In one embodiment, volume comprises a measurement of a muscle group. In one embodiment, volume comprises a measurement of the thigh muscle. In one embodiment, volume comprises a measurement of subcutaneous fat. In one embodiment, volume comprises a measurement of the visceral fat. In one embodiment, volume comprises a measurement of the liver fat. In one embodiment, volume comprises a measurement of the intramuscular fat. As used herein, the term "volume of at least one of the segments" refers to the amount of space occupied by one or more segments, e.g., those corresponding with an individual piece of the body. In one embodiment, the volume of at least one of the segments is the weight of the body, a part of the body, or parts of the body. In one embodiment, the volume of at least one of the segments is the total fat content of the torso.

In one embodiment, the body is a human body.

As used herein, the term "human body" refers to a mammal of the genus homo sapien. Common characteristics of a human body include bipedalism, opposable thumbs, and the basic anatomy of two legs, two arms, a torso, a neck, and a head.

In one embodiment of the methods disclosed herein, the images for the body comprise data chosen from radiography, nuclear emission, Magnetic Resonance Imaging (MRI), Ultrasound, Photoacoustic imaging, Thermography, Tomography, Echocardiography, or Functional near-infrared spectroscopy.

As used herein, the term "data" refers to certain, definite curated values. Data comprises many different formats and sources. In one embodiment, data is in DICOM (Digital Imaging and Communications in Medicine) files. In one embodiment, data is in ΝΙΠΊ volumes. In one embodiment, DICOM files are converted into ΝΙΠΊ volumes. In one embodiment, ΝΙΠΊ series are merged into a single whole-body volume. In one embodiment, data is in binary code. In one embodiment, data comprises information about the body. In one embodiment, data is collected from MRI scans. In one embodiment, data is stored in a cloud database platform. As used herein, the term "radiography" refers to an imaging technique using electromagnetic radiation to view the internal structure of a non-uniformly composed and opaque object, i.e., a non-transparent object of varying density and composition. In one embodiment, radiography is used to take an internal image of a human body. In one embodiment, radiography is used to take an internal image of a dog body.

As used herein, the term "nuclear emission" refers to using an emission spectra for generating an image. In one embodiment, an ion beam is used to focus on an object to fill the field of view of a special emission particle microscope system formatted with a single particle position sensitive detector (PSD). In one embodiment, nuclear emission is used to take an image of a body. In one embodiment, nuclear emission is used to take an image of a human body. In one embodiment, nuclear emission is used to take an internal image of a human body.

As used herein, the term "Magnetic Resonance Imaging" or "MRI" refers to using magnetic energy for generating a picture of the internal anatomy and the physiological processes of the body. MRI uses strong magnetic fields, radio waves, and field gradients to generate images of the inside of the body. Certain atomic nuclei absorb and emit radio frequency energy when placed in an external magnetic field. In one embodiment, atoms (for example H, C, P, Na, or other atoms on the periodic table of the elements) are used to generate a detectable radio-frequency signal that is received by antennas close to the anatomy being examined. Hydrogen atoms exist naturally in water and fat and thus an MRI scan can essentially map the location of water and fat in the body. In one embodiment, pulses of radio waves excite the nuclear spin energy transition, and magnetic field gradients localize the signal in space. By varying the parameters of the pulse sequence, different contrasts can be generated between tissues based on the relaxation properties of the hydrogen atoms therein. In one embodiment, MRI is used to form images of non-living objects. In one embodiment, MRI is used to form images of the body. In one embodiment, MRI is used to form images of a human body.

As used herein, the term "ultrasound" refers to using sound waves. In one embodiment, ultrasound is used for detecting objects and measuring distances. In one embodiment, a probe generates sound waves across an object and the reflected sound waves are recorded with a computer for generating an image. In one embodiment, an ultrasound is used for generating an image of a body. In one embodiment, an ultrasound is used for generating an image of a human body. As used herein, the term "photoacoustic imaging" refers to making a visual representation through the formation of sound waves following light absorption in a material sample. In one embodiment, photoacoustic imaging comprises non-ionizing laser pulses delivered into tissue of the body. In one embodiment, the delivered energy will be absorbed and converted into heat, leading to transient thermoelastic expansion and thus wideband (i.e. MHz) ultrasonic emission. In one embodiment, generated ultrasonic waves are detected by ultrasonic transducers and then analyzed to produce images. In one embodiment, photoacoustic imaging reveals physiologically specific optical absorption contrast. In one embodiment, two-dimensional images of the targeted areas are formed. In one embodiment, three-dimensional images of the targeted areas are formed. In one embodiment, images of the body are formed.

As used herein, the term "thermography" refers to detecting radiation in the long-infrared range of the electromagnetic spectrum and producing images of that radiation, called thermograms. In one embodiment, the long-infrared range is between 9,000-14,000 nanometers. In one embodiment, thermography comprises Infrared thermography (IRT). In one embodiment, thermography comprises thermal imaging. In one embodiment, thermography comprises thermal video. Infrared radiation is emitted by all objects, with a temperature above absolute zero based on the law of blackbody radiation, thermography allows one to view the environment with or without visible illumination. In one embodiment, the amount of radiation emitted by an object increases with temperature; therefore, thermography allows one to see variations in temperature. In one embodiment, when viewed through a thermal imaging camera, warm objects stand out well against cooler backgrounds. In one embodiment, humans and other warm-blooded bodies become easily visible against the environment, day or night.

As used herein, the term "tomography" refers to imaging by sections or sectioning, through the use of any kind of penetrating wave. Tomography is used in radiology, archaeology, biology, atmospheric science, geophysics, oceanography, plasma physics, materials science, astrophysics, quantum information, and other areas of science. In one embodiment, tomography comprises producing images based on mathematical procedure tomographic reconstruction. In one embodiment, tomography is X-ray scanning. In one embodiment, tomography is ultrasound imaging. In one embodiment, tomography produces images of the body. In one embodiment, tomography produces images of the human body. As used herein, the term "echocardiography" refers to using waves to investigate the heart. In one embodiment, echocardiography uses two-dimensional waves to create images of the heart. In one embodiment, echocardiography uses three-dimensional waves to create images of the heart. In one embodiment, echocardiography uses Doppler waves to create images of the heart. Echocardiography is often used to diagnose, manage, and follow-up patients with any suspected or known heart diseases. Echocardiography is one of the most widely used diagnostic tests in cardiology. In one embodiment, echocardiography provides the size and shape of the heart (internal chamber size quantification), pumping capacity, and the location and extent of any tissue damage. In one embodiment, echocardiography provides other estimates of heart function, such as a calculation of the cardiac output, ejection fraction, and diastolic function (how well the heart relaxes). In one embodiment, echocardiography detects cardiomyopathies, such as hypertrophic cardiomyopathy, dilated cardiomyopathy, and many others.

As used herein, the term "functional near-infrared spectroscopy" refers to neuroimaging technology offering a method of indirect and direct monitoring of brain activity. In one embodiment, functional near-infrared spectroscopy measures the changes in near-infrared light, allowing one to monitor blood flow in the front part of the brain. In one embodiment, functional near-infrared spectroscopy allows functional imaging of brain activity (or activation) through monitoring of blood oxygenation and blood volume in the prefrontal cortex. In one embodiment, functional near-infrared spectroscopy measures the changes in the concentration of oxy- and deoxy-haemoglobin (Hb) as well as the changes in the redox state of cytochrome-c-oxidase (Cyt-Ox) by their different specific spectra in the near- infrared range between 700-1000 nm. In one embodiment, functional near-infrared spectroscopy comprises functional near-infrared spectroscopy (fNIRS). In one embodiment, functional near-infrared spectroscopy comprises a functional near-infrared spectroscopy (fNIRS) sensor attached to the body's forehead and is either connected directly to a computer, or a portable computing device that records the body's data as it engages in specific tasks. In one embodiment, the methods disclosed herein comprise collecting Magnetic Resonance Imaging (MRI) data.

In one embodiment, the methods disclosed herein comprise transforming the plurality of two-dimensional images for the body into biologically relevant and anatomically specific volume-based measurements. As used herein, the term "transforming" refers to changing from one form to another. In one embodiment, transforming comprises changing DICOM to NlfTI (individual slabs). In one embodiment, transforming comprises utilizing a machine learning program. In one embodiment, transforming comprises using a cloud database platform. As used herein, the term "biologically relevant" refers to the importance to the life of a particular organism. In one embodiment, biologically relevant images comprise images showing cholesterol levels. In one embodiment, biologically relevant data comprises information about muscle development. In one embodiment, biologically relevant images comprise images displaying fat distribution.

As used herein, the term "anatomically specific volume-based measurement" refers to the amount of space within a specific structure of the body. In one embodiment, an anatomically specific volume-based measurement is the amount of fat in the torso. In one embodiment, an anatomically specific volume-based measurement is the tissue density of the thighs. In one embodiment, an anatomically specific volume-based measurement is the bone density of the leg. In one embodiment, an anatomically specific volume-based measurement is the size of the brain.

In one embodiment of the methods disclosed herein, assigning landmarks comprises aligning data from multiple different bodies.

As used herein, the term "aligning data" refers to performing point registration. In one embodiment, aligning data happens before image registration. In one embodiment, a rigid transformation (e.g., affine transformation) is found that best aligns the hips and shoulders. In one embodiment, the previous embodiment is used to initialize a non-rigid registration on the images of the torso.

As used herein, the term "multiple different bodies" refers to more than one body. In one embodiment, multiple different bodies is human bodies of different ages. In one embodiment, multiple different bodies is bodies of the male gender. In one embodiment, multiple different bodies is a collection of different bodies from one physical geographic location.

In one embodiment, the methods disclosed herein comprise identifying Random Forests or Random Ferns. As used herein, the term "Random Forests" refers to a machine learning algorithm for classification, regression and other tasks, operating by constructing a multitude of decision trees at training time and outputting the class that is the mode of the classes (classification) or mean prediction (regression) of the individual trees. In one embodiment Random Forests correct for decision trees' habit of overfitting to their training set. In one embodiment, Random Forests detect landmarks. In one embodiment, Random Forests detect bone joints. In one embodiment, Random Forests detect landmarks independently. In one embodiment, Random Forests are trained to predict one landmark, which only searches within a localized region. In one embodiment, Random Forests are used to manually segment organs/tissue. In one embodiment, Random Forests are trained by minimising mean squared errors (e.g., regression). In one embodiment, Random Forests are trained by maximising information gain (classification). In one embodiment, Random Forests are trained by randomly subsampling training data for each tree. In one embodiment, Random Forests are trained by randomly subsampling possible features at every node. In one embodiment, downsampling and/or upsampling is used within a process that is distinct from the detection process. In one embodiment, Random Forests produce pixelwise predictions for the full resolution of the image. In one embodiment, Random Forests produce pixelwise predictions for a downsampled version of the image. In one embodiment, the predictions are subsequently upsampled to the original image resolution. As used herein, the term "Random Ferns" refers to a machine learning algorithm for matching the same elements between two images of the same object. In one embodiment, Random Ferns recognizes certain objects. In one embodiment, Random Ferns traces objects on videos. In one embodiment, Random Ferns predict independent landmarks and parse the whole image at a lower cost than Random Forests. In one embodiment, a single set of Random Ferns predicts the whole spine of the body. In one embodiment, Random Ferns detects the spine curve of a body. In one embodiment, Random Ferns estimates the continuous curve following the contours of a spinal cord.

In one embodiment, the methods disclosed herein comprise comparing mean intensity of pairs of three-dimensional patches.

As used herein, the term "comparing" refers to determining the similarities and differences between two or more things. In one embodiment, comparing is between two bodies. In one embodiment, comparing is between two human bodies. In one embodiment, comparing comprises using a machine learning program. In one embodiment, comparing refers to juxtaposing two patches, such as patch A and patch B. For example, in one embodiment, comparing patch A and patch B means ordering the mean intensity of patch A and patch B. For example, patch A and patch B may be ordered where the mean intensity of patch A is greater than the mean intensity of patch B.

As used herein, the term "mean intensity" refers to the average intensity. In one embodiment, the mean intensity refers to the mean intensity of pixels. In one example the term "mean intensity" refers to the sum of all pixel values divided by the number of pixels. The term "mean intensity image" or "template" refers to an image in which the pixel values are chosen to be the mean intensity of the same pixel location in a set of images. In one embodiment, mean intensity is used to compare pairs of three-dimensional patches.

As used herein, the term "three-dimensional patch" refers to a section of the three- dimensional representation of the body. A three-dimensional patch allows one to closely investigate a piece of the body in fuller detail. In one embodiment, the methods disclosed herein comprise predicting three-dimensional offsets from current voxel.

As used herein, the term "predicting" means determining whether a future event will occur and/or how it will occur. In one embodiment, predicting is determining whether a three- dimensional representation of the body is accurate. In one embodiment, predicting is determining the location of segments of the body. In one embodiment, predicting refers to determining a value when the said value is unknown, for example, by using exogenous and/or endogenous information. In one embodiment, predicting comprises using a machine learning program. In one embodiment, predicting comprises a machine learning program assigning a value, e.g., score, label, or three-dimensional offset, based on what an algorithm has learned from the training data and based on the new data is it is currently observing.

As used herein, the term "three-dimensional offset" refers to a three dimensional section from the representation of the body moved to another location. In one embodiment, a three- dimensional offset also refers to a translation vector. In one embodiment, a three- dimensional offset is a landmark moved to another position. In one embodiment, a three- dimensional offset is a three-dimensional patch moved to another position. As used herein, the term "voxel" means a value on a grid in three dimensional space. In one embodiment, a voxel does not have its position/coordinates explicitly encoded along with its values. In one embodiment, a voxel's position is inferred based on its position relative to other voxels (i.e., its position in the data structure that makes up a single volumetric image).

As used herein, the term "current voxel" refers to the value on the grid used as the reference point in determining the position of other voxels. In one embodiment, the current voxel is used to identify all voxels associated with the arms and shoulders. In one embodiment, the current voxel is the voxel being considered at a specific time point of an algorithm looping over all the voxels. In one embodiment, the methods disclosed herein comprise a regression analysis.

As used herein, the term "regression analysis" refers to a prediction of properties based on unknown data. In one embodiment, regression analysis comprises a machine learning program. In one embodiment, the desired output consists of one or more continuous variables, e.g., X offset, Y offset, and Z offset in a three-dimensional representation.

In one embodiment, the methods disclosed herein comprise detecting each landmark independently. As used herein, the term "detected independently" refers to recognizing a feature without reliance upon other factors, e.g., a landmark. In one embodiment, Random Forests are trained to detect specific landmarks without reliance on information from other landmarks. In one embodiment, bone joints are detected independently. In one embodiment, the methods disclosed herein comprise implementing a spring-like shape model for enforcing global consistency.

As used herein, the term "implementing a spring-like shape model" means utilizing a rigid core for a surface mesh model and adding a new generalized spring for each mass. In one embodiment, the surface mesh model preserves its original geometric features such as volume and shape. Then, a shape matching approach updates the rigid core of the model dynamically so as to simulate global deformations. Finally, inverse dynamics technique are used to deal with the resulting deformations. In one embodiment, "implementing a spring-like shape model" means learning the variation in a relative position between the different parts of an object so that a score can be assigned to a set of possible configurations of parts. In one embodiment, the configuration with the highest score is selected as the most likely position of the object. See Fig. 1. as a non- limiting example of how to recreate a face.

As used herein, the term "enforcing global consistency" means causing uniformity among the body representation. Enforcing global consistency allows for accurate portrayal of the body and any modifications to the body images allows one to view the effect of on the entire body.

In one embodiment of the methods disclosed herein, registering comprises atlas-based segmentation.

As used herein, the term "atlas" refers to a specific model for a population of images with parameters that are learned from a training dataset. Images of a body can vary because of different body types and sizes, an atlas provides a way for accounting for these variations. In one embodiment, an atlas is used to align multiple images of the same body. In one embodiment, principal components analysis is used to match similar bodies for atlas based segmentation.

As used herein, the term "atlas-based segmentation" refers to a method of corrective separation through the use of a specific model for a population of images with parameters that are learned from a training dataset. In one embodiment, atlas based segmentation is the creation of a representation of a body through multiple images of a body.

In one embodiment, the methods disclosed herein comprise initializing atlas-based segmentation from detected landmarks.

As used herein, the term "initializing" means preparing for future actions. In one embodiment, initializing comprises segmenting a body. In one embodiment, initializing comprises detecting landmarks. In one embodiment, initializing comprises transferring data.

As used herein, the term "initializing atlas-based segmentation" refers to preparing a specific model for a population of images by training a dataset to set the parameters to correct segments. As used herein, the term "detected landmark" refers to previously marked or known reference of a thing. In one embodiment, a detected landmark is a labeled bone joint. In one embodiment, a detected landmark is an arm joint.

In one embodiment, the methods disclosed herein comprise non-rigid registration to an anatomical template.

As used herein, the term "anatomical template" refers to a pattern of the structure of a body. In one embodiment, an anatomical template is created from a plurality of three-dimensional representations of a body. In one embodiment, an anatomical template is created from a plurality of three-dimensional representations of a plurality of bodies.

In one embodiment of the methods disclosed herein, segments are chosen from fat, organs, bone, and muscle. As used herein, the term "fat" refers a biological material having both structural and metabolic functions. Fat is one of the three main macronutrients serving as an important foodstuff for many forms of life. In one embodiment, fat is a natural oily or greasy substance occurring in human bodies, especially when deposited as a layer under the skin or around certain organs. In one embodiment, fat is a triglyceride, an ester of three fatty acid chains and alcohol glycerol.

The terms "oil", "fat", and "lipid" are often confused and used interchangeably. "Oil" normally refers to a fat with short or unsaturated fatty acid chains and is liquid at ambient temperatures. "Fat" may specifically refer to fats that are solids at ambient temperatures. "Lipid" is the general term, as a lipid is not necessarily a triglyceride. Fats, like other lipids, are generally hydrophobic, and are soluble in organic solvents and insoluble in water. Fats and oils are categorized according to the number and bonding of the carbon atoms in the aliphatic chain. Fats that are saturated fats have no double bonds between the carbons in the chain. Unsaturated fats have one or more double bonded carbons in the chain. Nomenclature is based on the non-acid (non-carbonyl) end of the chain, called the omega end or the n-end.

Some oils and fats have multiple double bonds and are therefore called polyunsaturated fats. Unsaturated fats can be further divided into cis fats, which are the most common in nature, and trans fats, which are rare in nature. Unsaturated fats can be altered by reaction with hydrogen effected by a catalyst. This action, called hydrogenation, tends to break all the double bonds and makes a fully saturated fat. However, trans fats are generated during hydrogenation as contaminants created by an unwanted side reaction on the catalyst during partial hydrogenation.

In one embodiment, fat is the adipose, or fatty tissue, which serves as the body's means of storing metabolic energy over extended periods of time. Adipocytes (fat cells) store fat derived from the diet and from liver metabolism. Under energy stress these cells may degrade their stored fat to supply fatty acids and also glycerol to the circulation. These metabolic activities are regulated by several hormones (e.g., insulin, glucagon and epinephrine).

As used herein, the term "organ" refers to a collection of biological tissues joined in a structural unit to serve a common function. In one embodiment, organs are composed of main tissue, parenchyma, sporadic tissues, and stroma. The main tissue is unique for that specific organ. In one example, myocardium is the main tissue of the heart. In one embodiment, sporadic tissues include the nerves, blood vessels, and connective tissues.

In one embodiment, functionally related organs cooperate to form whole organ systems. In one embodiment, organs exist in biological organisms, in particular they are not restricted to animals, but also found in plants.

In one embodiment, a hollow organ is a visceral organ that forms a hollow tube or pouch, such as the stomach or intestine, or that includes a cavity, like the heart or urinary bladder.

As used herein, the term "bone" refers to a rigid collection of tissue constituting part of the vertebral skeleton. Bones support and protect the various organs of the body, produce red and white blood cells, store minerals, and also enable mobility as well as support for the body. In one embodiment, bone tissue is a type of dense connective tissue. Bones come in a variety of shapes and sizes and have a complex internal and external structure. Other types of tissue found in bones include, but are not limited to, marrow, endosteum, periosteum, nerves, blood vessels, and cartilage.

As used herein, the term "muscle" refers to a soft tissue found in a body. Muscle cells contain protein filaments, e.g., actin and myosin, sliding past one another producing a contraction that changes both the length and the shape of the cell. In one embodiment, muscles function to produce force and motion. In one embodiment, muscles are primarily responsible for maintaining and changing posture, locomotion, as well as movement of internal organs, such as the contraction of the heart, and the movement of food through the digestive system via peristalsis. In one embodiment, the muscle is skeletal/striated. In one embodiment, the muscle is cardiac. In one embodiment, the muscle is smooth.

In one embodiment of the methods disclosed herein, segmenting comprises multi-atlas segmentation.

As used herein, the term "multi-atlas segmentation" refers to a more than one specific model for a population of images with parameters that are learned from a training dataset for separation. In one embodiment, multi-atlas registration is used to represent an image as a deformed version of a template. In one example, there could be one template for a healthy population and one template for a diseased population. In one embodiment, the number of templates is determined by relying on image registration to transfer segmentation labels from pre-labeled atlases to a novel target image and applying label fusion to reduce errors produced by registration-based label transfer.

In one embodiment of the methods disclosed herein, segmenting comprises a weighted average of warped atlases.

As used herein, the term "weighted average" refers to a mean of values wherein some values contribute more to the final calculation than other values. When calculating a weighted average more preference may be assigned to a class of values than others. In one example, a class of atlases is assigned a higher value based on a number of factors, e.g., accuracy, uniformity, age, etc. As such, when an average of these atlases are calculated the specific class of atlases will have a higher effect on the overall average.

As used herein, the term "warped atlases" refers to a specific model for a population of images with parameters that are learned from a training dataset used for transforming the geometry of an image in order to superimpose one image on another. In one embodiment of the methods disclosed herein, segmenting comprises a patch-label transfer based on fat-water ratios.

As used herein, the term "patch-label transfer" refers to moving annotations between two images. In one embodiment, patch label transfer comprises selecting a pixel in a first image, extracting a patch centered on this pixel, and looking for the most similar patch in a second image by searching only in a local neighborhood. In one embodiment, the annotations assigned to the pixel in the first image is the annotation of the pixel in the second image that is the center of the best patch found during the matching process.

As used herein, the term "fat-water ratio" refers to a calculation to determine fat content based on the amount of H2O present relative to the amount of fat. In one embodiment, percentages are converted to relative percentages to overcome inhomogeneities.

In one embodiment of the methods disclosed herein, a fat segment is intra-abdominal fat tissue. As used herein, the term "intra-abdominal fat tissue" refers to visceral tissue located in the abdominal cavity. In one embodiment, intra-abdominal fat tissue is semi-fluid.

In one embodiment of the methods disclosed herein, segmenting comprises isolating an abdominal cavity from a human body from below pelvis to above liver.

As used herein, the term "abdominal cavity" refers to the empty space within the body left between the organs. In one embodiment, the abdominal cavity is a part of the abdominopelvic cavity, located below the thoracic cavity and above the pelvic cavity. In one embodiment, the abdominal cavity has a domed-shaped roof. In one embodiment, the abdominal cavity comprises a thoracic diaphragm.

As used herein, the term "pelvis" refers to either the lower area of the trunk of a body or the bones occupying that area. In one embodiment, the lower area of the trunk of a body is referred to as the pelvic region. In one embodiment, the bones within the lower trunk of a body is called the bony pelvis or pelvic skeleton.

As used herein, the term "liver" refers to an organ of vertebrates and some other animals possessing a wide range of functions, including detoxification of various metabolites, protein synthesis, and the production of biochemicals necessary for digestion. In one embodiment, the liver is located in the upper right quadrant of the abdomen and below the diaphragm in human bodies.

In one embodiment, the methods disclosed herein comprise spine and torso segmentation. As used herein, the term "spine" refers to part of an axial skeleton. The vertebral column is often the defining characteristic of a vertebrate, in which the notochord (a flexible rod of uniform composition) found in all chordates has been replaced by a segmented series of bones, i.e., vertebrae separated by intervertebral discs. The vertebral column houses the spinal canal, a cavity that encloses and protects the spinal cord.

As used herein, the term "torso" refers to a central part of a body, from which the neck and limbs extend. In one embodiment, the torso includes the thorax and the abdomen. In one embodiment, most critical organs are housed within the torso. In one embodiment, the torso harbours the pectoral muscles. In one embodiment, the torso harbours the abdominal muscles. In one embodiment, the torso harbours the lateral muscles. In one embodiment, the torso harbours the epaxial muscles.

In one embodiment, the methods disclosed herein comprise removing skeletal attributes from the three-dimensional representation of the body.

As used herein, the term "removing" refers to eliminating something from a defined collection of things.

As used herein, the term "skeletal attribute" refers to a trait or characteristic based on the bones of an organism. In one embodiment, a skeletal attribute is the bone density. In one embodiment, a skeletal attribute is the bone size.

As used herein, the term "removing skeletal attribute" refers to eliminating a trait or characteristic based on the bones of an organism. In one embodiment, the bone density is removed from a three-dimensional representation of a body. In one embodiment, the bone size is removed from an image of a body. In one embodiment, the entire skeleton is removed from an image of the body.

In one embodiment, the methods disclosed herein comprise determining a fat fraction and a fc-T2 map. As used herein, the term "fat fraction" refers a fraction of fat within a particular mass, for example a measurement of the percentage of fat in the liver. In one embodiment, fat fraction is based on magnetic resonance (MR), the fraction of the liver MR signal attributable to liver fat. In one embodiment, fat fraction is based on proton density, the fraction of the liver proton density attributable to liver fat.

As used herein, the term "fc-T2 map" refers to a fat-corrected image of estimated T2 values. As used herein, the term "T2-weighted MRI" refers to the T2 relaxation within the progressive dephasing of spinning dipoles following the 90° pulse as seen in a spin-echo sequence due to tissue-particular characteristics. In one specific example, tissue-particular characteristics affect the rate of movement of protons, for example those protons found in water molecules. This is alternatively known as spin-spin relaxation.

In one embodiment, the methods disclosed herein comprise presenting a concrete representation of a physical attribute of the body chosen from physical abdominal subcutaneous fat, visceral fat, upper-leg muscle (left/right), iliopsoas muscle (left/right), torso muscle including chest, back, shoulders (left/right), and abdomen; volume of fat in milliliters; Subcutaneous fat; adipose tissue in the hypodermis region below the skin for torso, legs, and head.

As used herein, the term "presenting a concrete representation" refers to showing a definite visual representation of the body. In one embodiment, presenting a concrete representation comprises showing a three-dimensional image of the body. In one embodiment, presenting a concrete representation comprises assigning landmarks to images of the body.

As used herein, the term "physical attribute" refers to a trait or characteristic of the body relating to the mass. In one embodiment, a physical attribute is total body content. In one embodiment, a physical attribute is bone size.

As used herein, the term "abdominal subcutaneous fat" refers to fat underneath the skin and within the stomach region of the body. In one embodiment, abdominal subcutaneous fat surrounds organs.

As used herein, the term "visceral fat" refers to body fat stored within the abdominal cavity and around a number of important internal organs. In one embodiment, visceral fat surrounds the liver. In one embodiment, visceral fat surrounds the pancreas. In one embodiment, visceral fat surrounds the intestines.

As used herein, the term "upper-leg muscle (left/right)" refers to tissue that comprise the top half of a leg. Within the context of this disclosure, the leg is the left leg, right leg, or both. In one embodiment, the upper leg muscle is the bicep femoris. In one embodiment, the upper leg muscle the semimembranosus muscle. As used herein, the term "iliopsoas muscle (left/right)" refers to the combination of the psoas major and the iliacus at their inferior ends. In one embodiment, the iliopsoas muscle is distinct in the abdomen. In one embodiment, the iliopsoas muscle is indistinguishable in the thigh. Within the context of this disclosure, the iliopsoas muscle refers to iliopsoas muscle located on the right side of the body, iliopsoas muscle located on the left side of the body, or both.

As used herein, the term "torso muscle including chest" refers to the trunk of the body and the front upper area below the neck and above the abdomen. As used herein, the term "back" refers to the dorsal surface of the body below the shoulders and above the hips.

As used herein, the term "shoulder (left/right)" refers to the upper joint of the arm. Within the context of this disclosure, the shoulder may refer to the right shoulder, the left shoulder, or both.

As used herein, the term "abdomen" refers to the part of the body between the thorax and the pelvis, as well as the cavity of this part of the trunk containing the chief viscera. As used herein, the term "volume of fat in milliliters" refers to measuring the space occupied in cubic centimeters, aka milliliters, aka ml_.

As used herein, the term "subcutaneous fat" refers to fat found just beneath the skin of the body. In contrast, visceral fat is found in the peritoneal cavity and can be measured using body fat calipers to give a rough estimate of total body adiposity. In one embodiment, the body cavity is defined and all subcutaneous fat between the body cavity and the boundary of the body is estimated.

As used herein, the term "adipose tissue" refers to the loose connective tissue composed of adipocytes. In one embodiment, adipose tissue is used to store energy in the form of fat, although it also cushions and insulates the body.

As used herein, the term "hypodermis region" refers to the lowermost layer of the integumentary system in the body. The types of cells often found in the hypodermis are fibroblasts, adipose cells, and macrophages. The hypodermis is derived from the mesoderm and unlike the dermis is not derived from the dermatome region of the mesoderm. In one embodiment, the hypodermis regions is used for fat storage. In one embodiment, the hypodermis consists of loose connective tissue and lobules of fat. In one embodiment, the hypodermis contains larger blood vessels and nerves than those found in the dermis. In one embodiment of the methods disclosed herein, calculating comprises combining anatomical segmentation, morphological operations, cluster-based thresholding, and conditional random field (CRF) refinement.

As used herein, the term "combining" refers to merging pieces or things together. In one embodiment, combining comprises forming one unit. In one embodiment, combining comprises forming several larger units from many smaller units.

As used herein, the term "anatomical segmentation" refers to sorting pieces of the body into collections corresponding to particular structures.

As used herein, the term "morphological operation" refers to the application of a structuring element to an input image and creating an output image of the same size. In one embodiment, the value of each pixel in an output image is based on a comparison of the corresponding pixel in the input image of surrounding pixels. In one embodiment, by choosing the size and shape of the neighborhood, i.e., surrounding pixels, a morphological operation is constructed that is sensitive to specific shapes in the input image.

In one embodiment, morphological operation is dilation. Dilation adds pixels to the boundaries of objects in an image. In one embodiment, morphological operation is erosion. Erosion removes pixels on object boundaries. In one embodiment, the number of pixels added or removed from the objects in an image depends on the size and shape of the structuring element used to process the image.

As used herein, the term "cluster-based thresholding" or "CBT" refers to a method of using multiple comparisons for correcting statistical maps in neuroimaging studies. CBT is useful because of its high sensitivity to weak and diffuse signals. In one embodiment, CBT is used for reducing the gray level image to a binary image.

As used herein, the term "conditional random field (CRF)" refers to a class of statistical modelling method often applied in pattern recognition and machine learning, where they are used for structured prediction. In one embodiment, CRF is a type of discriminative undirected probabilistic graphical model. In one embodiment, CRF is used for labeling or parsing sequential data, such as natural language text or biological sequences and in computer vision. In one embodiment, CRF is used for image denoising, in particular for removing unwanted noise from the segmentation labels. In one embodiment, the methods disclosed herein comprise calculating volume of tissue for at least one muscle group.

As used herein, the term "volume of tissue" refers to the measurement of space a distinct type of material consisting of specialized cells and their products. In one embodiment, the volume of fat tissue is measured. In one embodiment, the volume of bone tissue is measured.

As used herein, the term "muscle group" refers to a classification of a band or bundle of fibrous tissue in the body having the ability to contract, produce movement, or maintain the position of parts of the body. In one embodiment, the muscle group is the pectoral. In one embodiment, the muscle group is the tricep. In one embodiment, the muscle group is the quadricep. In one embodiment, the muscle group is the calf. In one embodiment, the muscle group is the hamstring. In one embodiment, the muscle group is the forearm. In one embodiment, the methods disclosed herein comprise exposing left-right imbalances in muscle volume.

As used herein, the term "left-right imbalances" refers to disproportionate measurements of either side of the body. In one embodiment, the body is bisected along the vertical axis. In one embodiment, the left-right imbalances is the muscle volume.

As used herein, the term "exposing left-right imbalances" refers to detecting disproportionate measurements of either side of the body. In one embodiment, exposing left-right imbalances is an indication of an abnormality.

As used herein, the term "muscle volume" refers to the amount of space a band or bundle of fibrous tissue in the body occupies.

In one embodiment, the methods disclosed herein comprise isolating an image of grey matter within a brain. As used herein, the term "grey matter within a brain" refers to a major component of the central nervous system, consisting of neuronal cell bodies, neuropil (dendrites and myelinated as well as unmyelinated axons), glial cells (astroglia and oligodendrocytes), synapses, and capillaries. Grey matter is distinguished from white matter, in that grey matters contains numerous cell bodies and relatively few myelinated axons, while white matter contains relatively very few cell bodies and is composed chiefly of long-range myelinated axon tracts. The colour difference arises mainly from the whiteness of myelin. In living tissue, grey matter has a very light grey colour with yellowish or pinkish hues, which come from capillary blood vessels and neuronal cell bodies. In one embodiment, the methods disclosed herein comprise isolating an image of white matter within a brain.

As used herein, the term "white matter within a brain" refers to areas of the central nervous system mainly composed of myelinated axons called tracts. White matter actively affects learning and brain functions, modulating the distribution of action potentials, acting as a relay and coordinating communication between different brain regions. White matter is named for its relatively light appearance resulting from the lipid content of myelin. However, the tissue of the freshly cut brain appears pinkish white to the naked eye because myelin is composed largely of lipid tissue veined with capillaries. White matter's white color in prepared specimens is due to its usual preservation in formaldehyde.

In one embodiment, the methods disclosed herein comprise isolating an image of cerebrospinal fluid. As used herein, the term "cerebrospinal fluid" refers to a clear, colorless body fluid found in the brain and spine. In one embodiment, cerebrospinal fluid is produced in the choroid plexuses of the ventricles of the brain. In one embodiment, cerebrospinal fluid acts as a cushion or buffer for the brain's cortex, providing basic mechanical and immunological protection to the brain inside the skull. In one embodiment, cerebrospinal fluid occupies the subarachnoid space (between the arachnoid mater and the pia mater) and the ventricular system around and inside the brain and spinal cord.

In one embodiment, for the methods disclosed herein, calculating comprises iterative bias field correction and segmentation. As used herein, the term "iterative bias field correction" refers to adjusting values based on repeated calculations. Iterative bias field correction is important in order to produce an accurate three-dimensional representation.

Within the context of this disclosure, the process steps or procedures can be executed by a human, operator, machine, and/or any combination thereof. For example, any combination of the following non-limiting exemplary steps could be executed by a human, operator, machine, and/or any combination thereof:

1. Converting DICO to ΝΙΠΊ (individual slabs)

2. Reassembling NlfTI slabs into full volume

3. Detecting/correcting cardiac artifacts

4. Estimating spinal cord

5. Estimating bone-joint landmarks

6. Detecting and segmenting individual vertebrae

7. Upsampling manual segmentations using random forests with geodesic features

8. Converting fat and water signal intensities to relative percentages

9. Estimating body mask

10. Computing "feature vector"

1 1 . Detecting and segmenting male genitalia

12. Detecting and segmenting arms

13. Estimating boundary between the legs

14. Partitioning the body into anatomical regions

15. Segmenting the lungs and trachea

16. Segmenting the iliopsoas muscles

17. Segmenting the torso muscles (chest, back, abdomen and shoulders)

18. Detecting and segmenting the breasts

19. Segment ng the major leg muscles (upper and lower, left and right)

20. Segment ng the major leg bones (femur and tibia, left and right)

21 . Segment ng pelvic bone and iliacus muscles (left and right)

22. Segment ng kidneys

23. Segment ng liver

24. Segment ng ribcage

25. Segment ng subcutaneous fat

26. Segment ng visceral fat

27. Segment ng internal thigh fat Within the context of this disclosure, "Converting DICOM to NlfTI (individual slabs)" refers to assembling DICOM files together by series and converting into NlfTI volumes.

Within the context of this disclosure, "Reassembling NlfTI slabs into full volume" refers to merging all NlfTI series into a single whole-body volume. As well as automating and implementing fat-water swap detection/correction using only the subject's scan.

Within the context of this disclosure, "Detecting/correcting cardiac artifacts" refers to detecting motion-based cardiac artifacts and removing them to improve the quality of tissue segmentations.

Within the context of this disclosure, "Estimating spinal cord" refers to estimating a continuous curve following the contour of the spinal cord using random ferns. The location of the curve is posterior to and in between the spines of an individual vertebrae. This curve is used to exclude unwanted fat/muscle tissue when defining the abdominopelvic cavity for visceral fat segmentation.

Within the context of this disclosure, "Estimating bone-joint landmarks" refers to detecting the major bone joints (shoulders, hips, knees, ankles) using training data and machine learning techniques. The estimated locations form a coordinate system is unique to each subject and allows anatomically-specific partitions.

Within the context of this disclosure, "Detecting and segmenting individual vertebrae" refers to combining training datasets in a machine-learning framework to detect the centroids of the vertebrae (L5 - T8). The results are used to exclude fatty tissue from the visceral fat segmentation.

Within the context of this disclosure, "Upsampling manual segmentations using random forests with geodesic features" refers to manually generated segmented organs/tissue, obtained in a downsampled space and upsampling them to full resolution using an interpolator that is guided by the signal intensities of the data.

Within the context of this disclosure, "Converting fat and water signal intensities to relative percentages" refers to overcoming the fat and water signal intensity discrepancies caused by inhomogeneities by converting them to relative percentages. They are also used to estimate the volume estimates related to fat and non-fat tissues. Within the context of this disclosure, "Estimating body mask" refers to producing a binary mask including only tissue associated with the subject's body and separating the subject from air, the scanner table, etc.

Within the context of this disclosure, "Computing 'feature vector'" refers to applying Principal components analysis (PCA) to each subject's body composition above the hips. This value is used to determine which subject's in the database are most similar to a new subject for atlas-based segmentation.

Within the context of this disclosure, "Detecting and segmenting male genitalia" refers to defining a search volume in a region based on the hip landmarks and assuming the genitals are the only area where fat is not located close to the body surface. The genitals are then excluded from the fat segmentation routines.

Within the context of this disclosure, "Detecting and segmenting arms" refers to transforming the coordinate system in order to identify all voxels associated with the arms and shoulders. The arms and some of the shoulder tissue are then excluded from quantitative analysis.

Within the context of this disclosure, "Estimating boundary between the legs" refers to splitting the whole body into left and right components with particular attention to separating the legs.

Within the context of this disclosure, "Partitioning the body into anatomical regions" refers to defining four major components: head, torso, left leg and right leg.

Within the context of this disclosure, "Segmenting the lungs and trachea" refers to using the lack of an MR signal to extract the lungs and trachea.

Within the context of this disclosure, "Segmenting the iliopsoas muscles" refers to atlas- based segmentation and refinement procedures applied to the iliopsoas muscles and manually generated training data.

Within the context of this disclosure, "Segmenting the torso muscles (chest, back, abdomen, and shoulders)" refers to atlas-based segmentation and refinement procedures applied to the torso muscles and manually generating training data. Within the context of this disclosure, "Detecting and segmenting the breasts" refers to removing the non-fat breast tissue from fat segmentations using the mask of breast tissue from the torso segmentation. The mask of breast tissue from the torso segmentation is derived from manual or atlas-based segmentation techniques. Atlas-based segmentation and refinement procedures are then applied to produce a final segmentation.

Within the context of this disclosure, "Segmenting the major leg muscles (upper and lower, left and right)" refers to generating an initial mask of the left and right legs using only the subject's data. Atlas-based segmentation and refinement are then applied to produce a final segmentation.

Within the context of this disclosure, "Segmenting the major leg bones (femur and tibia, left and right)" refers to initially segmenting the femur and tibia based on the initial leg segmentation and the detected landmarks of the bone joints. The method is based on region growing and geodesic distances. Atlas-based segmentation and refinement procedures are then applied to produce a final segmentation.

Within the context of this disclosure, "Segmenting pelvic bone and iliacus muscles (left and right)" refers to atlas-based segmentation and refinement procedures applied to the pelvic bone and iliacus muscles and manually generated training data.

Within the context of this disclosure, "Segmenting kidneys" refers to atlas-based segmentation and refinement procedures applied to the kidneys and manually generated training data. Within the context of this disclosure, "Segmenting liver" refers to atlas-based segmentation and refinement procedures applied to the liver and manually generated training data.

Within the context of this disclosure, "Segmenting ribcage" refers to estimating the position of a thin surface containing the ribs and using a rib shape model and registration on the fat percentages.

Within the context of this disclosure, "Segmenting subcutaneous fat" refers to first defining the body cavity and then estimating all fat tissue between the body cavity and boundary of the body. Within the context of this disclosure, "Segmenting visceral fat" refers to defining the abdominopelvic cavity and eliminating all other tissues and non-relevant organs. Then, estimating all fat tissue within the abdominopelvic cavity.

Within the context of this disclosure, "Segmenting internal thigh fat" refers to using the leg bone and subcutaneous fat segmentations for segmenting the remaining fat and muscle tissue in the upper legs. The midpoint between the hips and knees is estimated and a fixed region of muscle tissue is defined.

In one embodiment, 20 subjects are matched to a current body to provide atlases for each muscle group. The torso/iliopsoas atlases are generated by a human operator. Then, leg atlases are generated automatically. Atlas-based registration construct a probability mask. Refinement for torso-muscle segmentation are made using graph cuts (continuous max flow). Refinement of leg-muscle segmentation are made using conditional random field (dense CRF). There is no refinement of the iliopsoas segmentation, only thresholding.

In one embodiment, data from single RI is used. A human operator estimates the body cavity as well estimating the subcutaneous fat by excluding the body cavity. Refinement of both segmentations are done automatically by conditional random field (dense CRF). Unions and intersections of anatomy are used to estimate the abdominal cavity. Then the body cavity, pelvic mask, lungs, spine, upper legs, torso muscle, and iliopsoas muscle are estimated. Only the visceral fat is estimated in the abdominal cavity.

In one embodiment, thigh muscles are isolated. A human operator segments the subcutaneous fat from the thigh. A machine learning program automatically removes the skeletal structure. The fat fraction is calculated from a Dixon MRI. Mapping T2 and fc-T2 is from additional sequencing.

Although the disclosed invention has been described with reference to various exemplary embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the present invention. Those having skill in the art would recognize that various modifications to the exemplary embodiments may be made, without departing from the scope of the invention.

Moreover, it should be understood that various features and/or characteristics of differing embodiments herein may be combined with one another. It is therefore to be understood that numerous modifications may be made to the illustrative embodiments and that other arrangements may be devised without departing from the scope of the invention.

Furthermore, other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a scope and spirit being indicated by the claims.

Finally, it is noted that, as used in this specification and the appended claims, the singular forms "a," "an," and "the," include plural referents unless expressly and unequivocally limited to one referent, and vice versa. As used herein, the term "include" or "comprising" and its grammatical variants are intended to be non-limiting, such that recitation of an item or items is not to the exclusion of other like items that can be substituted or added to the recited item(s).

Claims

Claims We claim:

1. A method of measuring body volume, comprising:

Collecting a plurality of two-dimensional images for a body;

Processing the plurality of two-dimensional body images into a three-dimensional representation of the body;

Assigning landmarks to points within the three-dimensional representation of the body;

Registering the three-dimensional representation of the body;

Segmenting the three-dimensional representation of the body into segments; and

Calculating the volume of at least one of the segments.

2. The method of claim 1 , wherein the body is a human body.

3. The method of claim 1 , wherein the images for the body comprise data chosen from radiography, nuclear emission, Magnetic Resonance Imaging (MRI), Ultrasound, Photoacoustic imaging, Thermography, Tomography, Echocardiography, or Functional near-infrared spectroscopy.

4. The method of claim 1 , comprising collecting Magnetic Resonance Imaging (MRI) data.

5. The method of claim 1 , comprising transforming the plurality of two-dimensional images for the body into biologically relevant and anatomically specific volume-based measurements.

6. The method of claim 1 , wherein assigning landmarks comprises aligning data from multiple different bodies.

7. The method of claim 1 , comprising identifying Random Forests or Random Ferns.

8. The method of claim 1 , comprising comparing mean intensity of pairs of three- dimensional patches.

9. The method of claim 1 , comprising predict three-dimensional offsets from current voxel.

10. The method of claim 9, comprising a regression analysis.

1 1 . The method of claim 1 , wherein each landmark is detected independently.

12. The method of claim 1 , comprising a implementing a spring-like shape model for enforcing global consistency.

13. The method of claim 1 , wherein registering comprises atlas-based segmentation.

14. The method of claim 13, comprising initializing atlas-based segmentation from detected landmarks.

15. The method of claim 1 , comprising non-rigid registration to an anatomical template.

16. The method of claim 1 , where the segments are chosen from fat, organs, bone, and muscle.

17. The method of claim 1 , wherein segmenting comprises multi-atlas segmentation.

18. The method of claim 1 , wherein segmenting comprises a weighted average of warped atlases.

19. The method of claim 1 , wherein segmenting comprises a patch-label transfer based on fat-water ratios.

20. The method of claim 15, wherein the segments are fat segments of the intraabdominal fat tissues.

21 . The method of claim 1 , wherein segmenting comprises isolating an abdominal cavity from a human body from below pelvis to above liver.

22. The method of claim 21 , comprising spine and torso segmentation.

23. The method of claim 1 , comprising removing skeletal attributes from the three- dimensional representation of the body

24. The method of claim 1 comprising determining a fat fraction and a fc-T2 map.

25. The method of claim 1 , comprising presenting a concrete representation of a physical attribute of the body chosen from physical abdominal subcutaneous fat, visceral fat, upper-leg muscle (left/right), iliopsoas muscle (left/right), torso muscle including chest, back, shoulders, and abdomen (left/right); volume of fat in milliliters; Subcutaneous fat; adipose tissue in the hypodermis region below the skin for torso, legs, and head;

26. The method of claim 1 , wherein calculating comprises combining anatomical segmentation, morphological operations, cluster-based thresholding, and conditional random field (CRF) refinement.

27. The method of claim 1 , comprising calculating volume of tissue for at least one muscle group.

28. The method of claim 1 , comprising exposing left-right imbalances in muscle volume.

29. The method of claim 1 , comprising isolating an image of grey matter within a brain.

30. The method of claim 1 , comprising isolating an image of white matter within a brain.

31 . The method of claim 1 , comprising isolating an image of cerebrospinal fluid.

32. The method of claim 1 , wherein calculating comprises iterative bias field correction and segmentation.