Systems, devices and methods for spinal analysis

Abstract

Embodiments include exemplary systems, methods, and computer-accessible media for analysis of anatomical images to assess stenosis and disc degeneration. In some embodiments, the systems, devices, and methods described herein include selecting a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, identifying one or more anatomical parts in the ROI, determining one or more parameters of the one or more anatomical parts, and assessing severity of spinal deformity based on the one or more parameters.

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Application No. 63 / 312,678, filed February 22, 2022, entitled "SYSTEMS, DEVICES, AND METHODS FOR SPINAL ANALYSIS," the entire disclosure of which is incorporated herein by reference.

[0002] The present disclosure relates generally to systems, devices and methods for spinal analysis, and specifically to analysis of anatomical image data for spinal deformity assessment. [Background technology]

[0003] The human spinal column is a complex system of bones and soft tissue structures. The spine, which forms part of the spinal column, serves as the central support structure of the body and is composed of many individual bones known as vertebrae. Intervertebral discs are located between adjacent vertebrae to provide support and cushioning between the vertebrae. The vertebrae and discs, along with other soft tissue structures in their vicinity (e.g., ligaments, nervous system structures, etc.), form the spinal column. Each disc is composed of a series of dense collagen layers called the annulus fibrosus and a hydrogel containing proteoglycans and water called the nucleus. Dehydration of the disc (e.g., dehydration of the annulus fibrosus and / or dehydration of the nucleus) due to age and other factors can lead to a loss of structural and functional integrity of the disc, referred to as "disc degeneration." Nerve roots exit the spine through openings between two adjacent vertebrae. The openings are called the intervertebral foramina. Abnormal narrowing of the intervertebral foramina can lead to spinal stenosis. Tissue and / or bone changes can lead to herniated discs, osteophyte formation, and other spinal deformities such as spondylolisthesis.

[0004] Spinal deformities or diseases can take many forms. Two common ones are spinal stenosis and degenerative disc disease. Spinal stenosis involves narrowing of the central canal of the spine. One example of spinal stenosis is degenerative lumbar spinal stenosis (DLSS). DLSS is a major cause of back pain and one of the most common indications for spinal surgery. DLSS has a high prevalence in the working age population, especially in men aged 40-50 years, and can cause high medical and / or social costs. Disc degeneration occurs as a result of changes in an individual's intervertebral discs. Degenerative disc disease can also cause back pain, which affects young to middle-aged people with a peak incidence at about age 40. Disc degeneration can also result in sciatica.

[0005] FIG. 11 illustrates the progression of degenerative disc disease. Typically, disc degeneration is a cascade of events resulting from loss of hydration of the disc, which can result in changes in the mechanical strain of the nucleus and annulus fibrosus, as well as a variety of other consequences, including loss of disc height (e.g., narrowing or thinning of the disc), which can narrow the foramen where the spinal nerve exits the spine, disc bulging or herniation, which also narrows the space for the spinal nerve, and other instabilities or deformations. When such instability occurs, the body can respond to the instability by growing bone across the joint, in a process known as "autofocus." In particular, the body can grow osteophytes or osteophytes around the disc and / or facet joint. Osteophyte growth can further compound the adverse effects of disc degeneration.

[0006] Spinal evaluation may allow for the detection of various spinal conditions and / or spinal deformities. Detecting and treating spinal conditions and / or spinal deformities at an early stage (e.g., when symptoms are mild) may prevent the condition from becoming severe. When performing a spinal evaluation, it may be important to identify specific locations, geometric limitations, and other parameters within the spine. These parameters may indicate the type of spinal condition and / or spinal deformity as well as its location within the spine. Analyzing image data of a person's anatomy may help identify existing spinal conditions and / or spinal deformities. Some existing imaging systems include, for example, computed tomography (CT), magnetic resonance imaging (MRI), x-ray, ultrasound, and fluoroscopy systems. Such imaging is widely utilized for both initial diagnosis and follow-up evaluations in both conservative and surgical treatment pathways.

[0007] Conventional X-rays and CT are common methods for obtaining information of a patient's anatomy, including, for example, the patient's spine. Conventional X-rays involve directing high-energy electromagnetic radiation at the patient's body and capturing the resulting two-dimensional (2D) X-ray profile on a film or plate. However, X-ray imaging can expose the patient to high levels of radiation. Analysis of X-rays can also be subjective based on the physician's training and experience. Currently, there is no autonomous way to objectively analyze X-rays. As such, making the necessary measurements on X-rays can be time consuming and subject to user error. The lack of an autonomous way to analyze X-rays also makes it difficult to objectively compare a patient's X-rays over time, for example to track a patient's progress. Due to these limitations, it is currently not possible to reliably predict a particular outcome based on X-ray imaging. Also, it is currently not possible to obtain the necessary measurements in an autonomous and / or consistent manner that ensures the reliability and reproducibility of such measurements.

[0008] CT involves obtaining 3D image data of a patient's anatomy using a controlled amount of X-ray radiation. Existing CT systems may include a rotating gantry with an X-ray tube mounted on one side and an arc-shaped detector mounted on the other side. As the rotating frame rotates the X-ray tube and detector around the patient, the X-ray beam can be emitted in a fan shape. Each time the X-ray tube and detector rotate 360° and the X-rays pass through the patient's body, an image of a thin section of the patient's anatomy can be obtained. During each rotation, the detector can record approximately 1,000 images or profiles of the expanded X-ray beam. Each profile can then be reconstructed by a dedicated computer into a 3D image of the scanned section. Thus, CT systems use multiple 2D CT scans or batches of X-rays to build a 3D image of the patient's anatomy. The speed of the gantry rotation, along with the slice thickness, contributes to the accuracy and / or usefulness of the final image. Commonly used intraoperative CT imaging systems have various settings that allow control of the radiation dose. In certain scenarios, high dose settings may be selected to ensure adequate visualization of anatomical structures. The downside is increased radiation exposure to the patient. The effective dose from a diagnostic CT procedure is typically estimated to be in the range of 1-10 millisieverts (mSv). Such high doses may result in increased risk of cancer and other health conditions. Therefore, low dose settings are selected for CT scans whenever possible to minimize radiation exposure and the associated risk of developing cancer. However, low dose settings may affect the quality of image data available to the surgeon.

[0009] MRI imaging systems work by forming a strong magnetic field around the area to be imaged. In most medical applications, protons (e.g., hydrogen atoms) in tissues, including water molecules, generate signals that are processed to form an image of the body. First, energy from an oscillating magnetic field is applied briefly to the patient at an appropriate resonant frequency. The excited hydrogen atoms emit a radio frequency (RF) signal, which is measured by an RF system. The RF signal may be made to encode location information by varying the main magnetic field using gradient coils. These coils are rapidly switched on and off, thus creating the characteristic repetitive noise of MRI scans. The contrast between different tissues may be determined by the rate at which the excited atoms return to their equilibrium state. In some cases, exogenous contrast agents may be administered intravenously, orally, or intra-articularly to further promote differentiation between different tissues. The main components of an MRI imaging system are a main magnet to polarize the tissues, shim coils to correct for inhomogeneities in the main magnetic field, a gradient system to localize the magnetic resonance (MR) signal, and an RF system to excite the tissues and detect the resulting signals. Different magnetic field strengths may be used in MRI imaging. The most common strengths are 0.3 T, 1.5 T and 3 T. The higher the strength, the better the image quality. For example, a field strength of 0.3 T produces lower quality imaging than a field strength of 1.5 T.

[0010] Currently, there is also no autonomous method to objectively analyze MRI images, and analysis of such images is dependent on the physician's training and experience. Furthermore, due to technical limitations, diagnostic MRI protocols provide a limited number of slices of the target region, which forces the physician to piece together anatomical information from available axial, sagittal, and / or coronal scans of the patient's anatomy. Existing systems also lack a reliable method to easily and autonomously compare a patient's MRI images to a larger database of MRI images. Such a comparison can allow the physician to obtain additional information regarding the severity of the patient's condition. Existing systems also lack the ability to autonomously compare a patient's current MRI images to the patient's past images. Furthermore, it is currently not possible to screen a patient's MRI images for spinal cord compression, fractures, tumors, infections, among other conditions. Such limitations make it difficult, if not impossible, to make treatment recommendations based on patient MRI images that would provide a high degree of confidence in the treatment outcome.

[0011] Even with these limitations, MRI is typically the imaging of choice for evaluating spinal stenosis due to its ability to image soft tissue detail. MRI also provides anatomical information that allows radiologists to identify the location, etiology, and severity of nerve root compression and report those findings. However, interpretation of MRI of spinal anatomy can be time consuming, especially when advanced multilevel degeneration is present. Although image-based grading systems such as the Lee grading or Pfirrmann grading systems exist, inter-reader variability is high, even among experts, reducing the perceived value of reader reporting. Therefore, a solution that provides accurate and consistent interpretation of imaging data of stenosis that can be applied on a large scale would be of high clinical utility.

[0012] Similarly, to evaluate the pathology and degeneration of the intervertebral disc, MRI is commonly used due to the lack of radiation, multiplanar imaging capability, high spinal soft tissue contrast, and precise localization of disc changes. In some cases, anterior-posterior (AP) and lateral views of X-rays can also help visualize the gross anatomical changes of the intervertebral disc. However, MRI is the standard imaging modality for detecting intervertebral disc lesions, because its aforementioned advantages present high clinical value in providing clinicians with information about intervertebral disc degeneration.

[0013] Existing systems pose diagnostic challenges for physicians due to poor quality images and lack of reliable and / or reproducible image analysis. Such limitations can make it difficult to properly identify important landmarks and take measurements. This can make it difficult to accurately perform spinal assessments to detect spinal conditions and / or spinal deformities. Accordingly, additional systems, devices, and methods for spinal analysis to detect spinal conditions and / or spinal deformities may be desirable. In particular, solutions that provide accurate and consistent interpretation of imaging data, especially MRI imaging data, are desirable. Summary of the Invention

[0014] The systems, devices, and methods described herein relate to the analysis of anatomical images and the identification of anatomical components and / or structures. In some embodiments, the systems, devices, and methods described herein relate to the identification of spinal conditions and / or spinal deformities.

[0015] In some embodiments, the method includes selecting a region of interest (ROI) within image data of a three-dimensional (3D) volume of multiple vertebrae of a spine, the ROI including image data of one or more vertebrae from the multiple vertebrae and tissue structures surrounding the one or more vertebrae; identifying the spinal cord or the thecal sac in the ROI; determining one or more parameters associated with the spinal cord or the thecal sac in the ROI; and determining a severity of spinal stenosis in the ROI based on the one or more parameters associated with the spinal cord or the thecal sac.

[0016] In some embodiments, the device includes a memory and a processor operably coupled to the memory, the processor configured to: select a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae; identify a spinal cord or a thecal sac in the ROI; determine one or more parameters associated with the spinal cord or thecal sac in the ROI; and determine a severity of spinal stenosis in the ROI based on the one or more parameters associated with the spinal cord or thecal sac.

[0017] In some embodiments, a non-transitory processor-readable medium storing code representing instructions to be executed by a processor, the code including code that causes the processor to select a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae, identify a spinal cord or a thecal sac in the ROI, determine one or more parameters associated with the spinal cord or thecal sac in the ROI, and determine a severity of spinal stenosis in the ROI based on the one or more parameters associated with the spinal cord or thecal sac.

[0018] In some embodiments, the method includes selecting a region of interest (ROI) within image data of a three-dimensional (3D) volume of multiple vertebrae of a spine, the ROI including image data of one or more vertebrae from the multiple vertebrae and tissue structures surrounding the one or more vertebrae; identifying one or more nerve roots in the ROI; tracing the one or more nerve roots from a lateral recess of the one or more vertebrae to a vertebral foramen to identify one or more regions including a nerve root discontinuity or a stenotic portion of the nerve root; and determining a severity or location of spinal stenosis in the ROI based on the one or more regions.

[0019] In some embodiments, the device includes a memory and a processor operatively coupled to the memory and configured to: select a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of the spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae; identify one or more nerve roots in the ROI; track the one or more nerve roots from a lateral recess of the one or more vertebrae to a vertebral foramen to identify one or more regions including a nerve root discontinuity or a stenotic portion of the nerve root; and determine a severity or location of spinal stenosis in the ROI based on the one or more regions.

[0020] In some embodiments, a non-transitory processor-readable medium storing code representing instructions to be executed by a processor, the code including: causing the processor to select a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae; identify one or more nerve roots in the ROI; track one or more nerve roots from a lateral recess of one or more vertebrae to a vertebral foramen to identify one or more regions including a nerve root discontinuity or a stenotic portion of the nerve root; and determine a severity or location of spinal stenosis in the ROI based on the one or more regions.

[0021] In some embodiments, the method includes selecting a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae; identifying an intervertebral disc in the ROI; determining one or more parameters of an annulus fibrosus and nucleus of the intervertebral disc; and determining a disc degeneration ratio based on the one or more parameters of the annulus fibrosus and nucleus of the intervertebral disc.

[0022] In some embodiments, the device includes a memory and a processor operably coupled to the memory, the processor configured to: select a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae; identify an intervertebral disc in the ROI; determine one or more parameters of an annulus fibrosus and a nucleus of the intervertebral disc; and determine a disc degeneration ratio based on the one or more parameters of the annulus fibrosus and the nucleus of the intervertebral disc.

[0023] In some embodiments, a non-transitory processor-readable medium storing code representing instructions to be executed by a processor, the code including code that causes the processor to select a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae, identify an intervertebral disc in the ROI, determine one or more parameters of an annulus fibrosus and a nucleus of the intervertebral disc, and determine a disc degeneration ratio based on the one or more parameters of the annulus fibrosus and the nucleus of the intervertebral disc. [Brief description of the drawings]

[0024] [Figure 1] FIG. 1 is a block diagram illustrating a configuration of a system for acquiring and analyzing anatomical images according to some embodiments.

[0025] [Diagram 2]1 illustrates a schematic diagram of an example of a computing device for spinal deformity diagnosis and / or spinal deformity assessment, according to some embodiments.

[0026] [Figure 3A] FIG. 1 is a schematic diagram illustrating a segmentation convolutional neural network according to some embodiments.

[0027] [Figure 3B] FIG. 1 is a schematic diagram illustrating an example of a model for performing level discrimination of spinal anatomy, according to some embodiments.

[0028] [Figure 4A] 1 is a flowchart illustrating a method for training and validating a model for performing spinal deformity assessment according to some embodiments.

[0029] [Figure 4B] 1 is a flowchart illustrating a training neural network training process for performing spinal deformity assessment according to some embodiments.

[0030] [Diagram 5] 1 is a flowchart illustrating a spinal deformity assessment according to some embodiments.

[0031] [Figure 6A] 1 is a flow chart of a level identification process according to some embodiments.

[0032] [Figure 6B] 1 is a flowchart of a spinal deformity assessment process according to some embodiments.

[0033] [Figure 7A] 1 is a flow chart of a stenosis assessment process according to some embodiments.

[0034] [Figure 7B]1 is a flowchart of a disc degeneration assessment process according to some embodiments.

[0035] [Figure 8] 1 illustrates an example of a report generated by the systems and methods described herein, according to some embodiments.

[0036] [Figure 9A] 1 shows an example of a report generated for a patient with a mild stenosis. [Figure 9B] 1 shows an example of a report generated for a patient with a mild stenosis. [Figure 9C] 1 shows an example of a report generated for a patient with a mild stenosis. [Figure 9D] 1 shows an example of a report generated for a patient with a mild stenosis. [Figure 9E] 1 shows an example of a report generated for a patient with a mild stenosis.

[0037] [Figure 10A] 1 shows an example of the diagnosis of severe stenosis in a patient.

[0038] [Figure 10B] An example of a patient diagnosed with no stenosis is shown.

[0039] [Figure 11] 1 is an illustration of the progression of spinal disc degeneration.

[0040] [Figure 12] 1 is a flow chart of a level identification process according to some embodiments.

[0041] [Figure 13A] 13A-13C visually illustrate the process of identifying and allocating levels to different MRI image sets according to some embodiments.

[0042] [Figure 13B]1 is a schematic visualization of combining MRI sagittal data with axial data, according to an embodiment.

[0043] [Figure 14A] 1 is an illustration of a spine under normal conditions and with spinal stenosis.

[0044] [Figure 14B] FIG. 1 is an illustration of different classifications of lumbar central canal stenosis based on the grading system developed by Lee et al. (the "Lee grading system").

[0045] [Figure 15] 1 is a flow chart of a stenosis assessment process according to some embodiments.

[0046] [Figure 16] 13A-13C are schematic diagrams illustrating a spinal stenosis analysis report according to some embodiments.

[0047] [Figure 17A] 13 provides an example of a spinal stenosis analysis report, according to an embodiment. [Figure 17B] 13 provides an example of a spinal stenosis analysis report, according to an embodiment. [Figure 17C] 13 provides an example of a spinal stenosis analysis report, according to an embodiment. [Figure 17D] 13 provides an example of a spinal stenosis analysis report, according to an embodiment.

[0048] [Figure 18A] 1 provides a detailed view of a graph correlating different classifications or grades of spinal stenosis with dural sac surface area, according to an embodiment. [Figure 18B] 1 provides a detailed view of a graph correlating different classifications or grades of spinal stenosis with dural sac surface area, according to an embodiment.

[0049] [Figure 19] 1 is an illustration of an intervertebral disc under normal conditions and with disc degeneration.

[0050] [Figure 20A] FIG. 1 is an illustration of different classifications of intervertebral disc degeneration based on the grading system developed by Pfirrmann et al. ("Pfirrmann grading system").

[0051] [Figure 20B] 1 is a flow diagram showing how the different classifications of the Pfirrmann grading system are assigned.

[0052] [Figure 21] 1 is a flowchart of a disc degeneration assessment process according to some embodiments.

[0053] [Figure 22] 13A-B are schematic diagrams illustrating a disc degeneration analysis report according to some embodiments.

[0054] [Figure 23] 13 provides an example of a disc degeneration analysis report, according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0055] 1. System Overview The systems, devices, and methods described herein relate to processing of a patient's anatomy, including the spine. Although the specific examples presented herein may relate generally to processing image data of the spine, one skilled in the art can appreciate that such systems, devices, and methods may be used to process image data of other parts of the patient's anatomy, including, for example, blood vessels, nerves, bones, and other soft and hard tissues near the brain, heart, or other regions of the patient's anatomy.

[0056] The systems, devices, and methods described herein may be suitable for processing a number of different types of image data, including x-ray, computed tomography (CT), magnetic resonance imaging (MRI), fluoroscopy, ultrasound, etc. In some embodiments, such systems, devices, and methods may process a single image type and / or view, while in other embodiments, such systems, devices, and methods may process multiple image types and / or views. In some embodiments, multiple image types and / or views may be combined to provide richer data about the patient's anatomy.

[0057] The systems, devices, and methods described herein may implement machine learning models for processing and / or analyzing image data related to a patient's anatomy. Such machine learning models may be configured to identify and distinguish different anatomical parts within the anatomy. In some embodiments, the machine learning models described herein may include neural networks, including deep neural networks having multiple layers between an input layer and an output layer. For example, one or more convolutional neural networks (CNNs) may be used to process the patient image data and generate a segmentation output that classifies different objects within the image data and / or identifies different levels of the spine within the image data. Suitable examples of segmentation models and their uses are described in U.S. Patent Application Publication No. 2019 / 0105009, published on November 11, 2019, entitled “Automatic Segmentation of Three-Dimensional Bone Structure Images,” U.S. Patent Application Publication No. 2020 / 0151507, published on May 14, 2020, entitled “Autonomous Segmentation of Three-Dimensional Nervous System Structures from Medical Imaging,” U.S. Patent Application Publication No. 2020 / 0151507, published on December 31, 2020, entitled “Autonomous Multidimensional Segmentation of Anatomical Structures in Three-Dimensional Medical Images,” and U.S. Patent Application Publication No. 2020 / 0152004, published on May 14, 2020, entitled “Autonomous Multidimensional Segmentation of Anatomical Structures in Three-Dimensional Medical Images.” No. 2020 / 0410687, U.S. Provisional Patent Application No. 63 / 187,777, filed May 12, 2021, entitled "Systems, Devices, and Methods for Segmentation of Anatomical Image Data," and PCT Patent Application No. PCT / US22 / 29000, filed May 12, 2022, entitled "Systems, Devices, and Methods for Segmentation of Anatomical Image Data," the disclosures of each of which are incorporated herein by reference. Suitable examples of methods for level identification are described in U.S. Patent Application Publication No. 2020 / 0327721, published on October 15, 2020, entitled "Autonomous Level Identification of Anatomical Bone Structures in 3D Medical Images," and U.S. Provisional Patent Application No. 63 / 256,306, filed on October 15, 2021, entitled "Level Identification in 3D Anatomical Images," the disclosures of each of which are incorporated herein by reference.Although the particular examples described herein use CNNs, it will be appreciated that other types of machine learning algorithms can be used to process patient image data, including, for example, support vector machines (SVMs), decision trees, k-nearest neighbors, and artificial neural networks (ANNs).

[0058] In some embodiments, the systems, devices, and methods described herein can implement machine learning models (e.g., spinal analysis models described further herein) to perform anatomical feature analysis and / or perform spinal deformity diagnosis or classification. For example, the systems, devices, and methods described herein can identify and distinguish between different anatomical parts in an anatomical structure, identify different levels of the spine, analyze anatomical features (e.g., discs, disc shapes and / or sizes, distances between adjacent vertebrae, etc.), perform spinal deformity assessment based on the analysis of anatomical features, associate the assessment with the identified level of the spine, and / or provide a report regarding the anatomical feature analysis or spinal deformity diagnosis.

[0059] 1 is a high-level block diagram illustrating a system 100 for processing image data of a patient's anatomy and / or providing image guidance to a physician during a surgical procedure, according to some embodiments. The system 100 can include a computing device 110, an imaging device 160, and optionally a surgical navigation system 170. In some embodiments, the computing device 110 can communicate with one or more imaging devices 160 and optionally one or more surgical navigation systems 170, for example, to perform segmentation of the patient's anatomy, perform level identification of the patient's anatomy, perform spinal analysis, detect spinal conditions and / or other spinal deformities, and / or provide digital guidance to a surgeon during a surgical procedure.

[0060] In some embodiments, the computing device 110 may be configured to perform segmentation of the anatomical image data to identify anatomical parts of interest. For example, the computing device 110 may be configured to generate a segmentation output that identifies different anatomical parts of interest. Additionally or alternatively, the computing device 110 may be configured to perform level identification of different regions of the spine. The computing device 110 may be configured to generate a level identification output, such as, for example, a level type (e.g., sacral, thoracic, lumbar, cervical), a vertebral level (ordinal identifier), or a pair or range of vertebral levels associated with a vertebra (and / or other nearby anatomical parts). Additionally or alternatively, the computing device 110 may be configured to perform anatomical feature analysis and / or spinal deformity assessment with respect to the various identified levels. For example, the computing device 110 may be configured to analyze anatomical parts of interest to determine a spinal condition and / or a spinal deformity. In some embodiments, the computing device 110 may be configured to associate the determined spinal condition and / or spinal deformity with one or more levels of the spine, for example based on the level identification output. In some embodiments, the computing device can be configured to generate a visual representation of the patient's anatomy and associate the level identification information, anatomical feature information, and / or spinal deformity diagnosis with different portions of the patient's anatomy. In some embodiments, the computing device 110 can be configured to generate a virtual representation of the patient's anatomy and / or surgical instruments, for example, to provide image guidance to a surgeon before and / or during a surgical procedure.

[0061] Computing device 110 may be implemented as a single computing device or across multiple computing devices that are connected to each other and / or to network 150. For example, computing device 110 may include one or more computing devices such as a server, a desktop computer, a laptop computer, a portable device, a database, etc. The different computing devices may be located remotely from other computing devices, located in a building near other computing devices, and / or include components integrated with other computing devices.

[0062] In some embodiments, the computing device 110 may be located on a server that is located remotely from the one or more imaging devices 160 and / or the surgical navigation system 170. For example, the imaging device 160 and the surgical navigation system 170 may be located in an operating room with the patient 180, while the computing device 110 may be located remotely but operably coupled (e.g., via the network 150) to the imaging device 160 and the surgical navigation system 170. In some embodiments, the computing device 110 may be incorporated into one or both of the imaging device 160 and the surgical navigation system 170. In some embodiments, the system 100 includes a single device that includes the functionality of the computing device 110, the one or more imaging devices 160, and the one or more surgical navigation systems 170, as described further herein.

[0063] In some embodiments, the computing device 110 may be located within a hospital or medical facility. The computing device 110 may be operably coupled to one or more databases associated with the hospital, such as a hospital database for storing patient information and the like. In some embodiments, the computing device 110 may be available to a physician (e.g., a surgeon) to perform evaluation of a patient's anatomical data (e.g., including the level data described herein), visualization of the patient's anatomical data, diagnosis, and / or planning of a surgical procedure. In some embodiments, the computing device 110 may be operably coupled to one or more other computing devices (e.g., physician workstations) within the hospital and may transmit the level output and / or other image processing output (e.g., via the network 150) to such computing devices to perform evaluation of a patient's anatomical data, visualization of the patient's anatomical data, diagnosis, and / or planning of a surgical procedure.

[0064] Network 150 may be any type of network (e.g., a local area network (LAN), a wide area network (WAN), a virtual network, a telecommunications network) implemented as a wired network and / or a wireless network and used to operatively couple the computing devices that comprise system 100. As shown in FIG. 1, connections may be defined between computing device 110 and any one of imaging device 160, surgical navigation system 170, and / or other computing devices (e.g., databases, servers, etc.). In some embodiments, computing device 110 may communicate with (e.g., transmit data to and / or receive data from) imaging device 160 and / or surgical navigation system 170 as well as network 150 via intermediate and / or alternative networks (not shown in FIG. 1). Such intermediate and / or alternative networks may be the same type and / or different type of networks as network 150. Each of the computing device 110, the imaging device 160, and the surgical navigation system 170 may be any type of device configured to transmit data over the network 150 to send and / or receive data from one or more of the other devices.

[0065] In some embodiments, imaging device 160 may refer to any device configured to image the anatomy of patient 180. In some embodiments, imaging device 160 may include one or more sensors for measuring signals generated by various imaging techniques. Imaging device 160 may use non-invasive techniques to image the patient's anatomy. Non-limiting examples of imaging devices include CT scanners, MRI scanners, X-ray devices, ultrasound devices, combinations thereof, and the like. Image data generated by imaging device 160 may be transmitted to any of the devices connected to network 150, including, for example, computing device 110. In some embodiments, image data generated by imaging device 160 may include a 2D image of the anatomy. In some embodiments, image data generated by imaging device 160 may include multiple 2D image scans that together provide image data of a 3D volume. Imaging device 160 can transmit image data to computing device 110, which can then perform segmentation of the patient's anatomy, perform level identification of the patient's anatomy, label different anatomical portions of interest within the patient's anatomy, perform spinal evaluation (e.g., spinal deformity diagnosis), detect spinal conditions and / or spinal deformities, and / or associate a spinal deformity diagnosis with the level identification information. Optionally, imaging device 160 can provide the image data to surgical navigation system 170, such that the surgical navigation system can generate one or more virtual representations of the patient's anatomy, for use in, for example, image-guided surgery.

[0066] The surgical navigation system 170 can be configured to provide image-guided surgery, for example, during a surgical procedure. For example, the surgical navigation system 170 can facilitate one or more of planning, visualization, and guidance during a surgical procedure. In some embodiments, the surgical navigation system 170 can include a tracking system for tracking the patient's anatomy, surgical instruments, implants, or other objects within the surgical field. In some embodiments, the surgical navigation system 170 can include an image generator for generating and displaying one or more virtual representations of the patient's anatomy and / or the surgical instruments, implants, or other objects within the surgical field to a physician or other healthcare provider (e.g., a surgeon). In some embodiments, the surgical navigation system 170 can be configured to present a 3D display, for example, via a 3D wearable device and / or a 3D projector or screen. In some embodiments, the surgical navigation system 170 can be configured to display the position and / or orientation of one or more surgical instruments and implants relative to pre-operative or intra-operative medical image data of the patient's anatomy. The image data may be provided, for example, by the imaging device 160, and the surgical navigation system 170 may use the image data to generate a virtual representation of one or more anatomical parts of interest along with position and / or orientation data associated with the surgical device. A suitable example of a surgical navigation system is described in U.S. Patent Application Publication No. 2019 / 0053851, published February 21, 2019, and incorporated herein by reference.

[0067] FIG. 2 illustrates an exemplary computing device 210 for surgical planning and / or navigation, according to some embodiments. The computing device 210 may be structurally and / or functionally similar to the computing device 110. Although a single computing device 210 is illustrated diagrammatically, it may be understood that the computing device 210 may be implemented as one or more computing devices. In some embodiments, the computing device 210 may be configured to segment a patient's anatomy, identify a level of the patient's spinal anatomy (e.g., patient 180), and / or perform anatomic feature analysis and / or spinal deformity analysis. The computing device 210 includes a processor 220, a memory 230, and one or more input / output interfaces 250.

[0068] The memory 230 may be, for example, a random access memory (RAM), a memory buffer, a hard drive, a database, an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), a read-only memory (ROM), etc. In some embodiments, the memory 230 stores instructions that cause the processor 220 to execute modules, processes, and / or functions related to segmentation 222, level identification 224, anatomical feature analysis 226, and deformity assessment 228. The memory 230 may store one or more of the segmentation model 232, the level identification model 234, the spine analysis model 252 (e.g., for stenosis assessment 252a, for disc degeneration assessment 252b), the anatomical part data 240, and / or the image data 242.

[0069] The segmentation model 232 may be a model or algorithm for performing image-based segmentation, thereby classifying or labeling different portions of anatomical image data. In some embodiments, the segmentation model 232 may include a machine learning model, such as, for example, a CNN model, an SVM model, etc. The segmentation model 232 may be implemented by the processor 220 to perform the segmentation 222. In some embodiments, the segmentation model 232 may be specific to a particular anatomical region, such as, for example, spinal anatomy, cardiac anatomy, etc. In some embodiments, the segmentation model 232 may be specific to a particular image type, such as, for example, X-ray, CT, MRI, etc.

[0070] The level identification model 234 may be a model or algorithm for identifying and / or labeling different levels of the vertebrae of the spine and / or other anatomical parts associated with those levels (e.g., nerves, discs, etc.). In some embodiments, the level identification model 234 may include a machine learning model, such as, for example, a CNN model, an SVM model, etc. The level identification model 234 may be implemented by the processor 220 to perform the level identification 224. In some embodiments, the level identification model 234 may be specific to a particular image type (e.g., X-ray, CT, MRI) and / or image view. For example, the level identification model 234 may include different models for identifying levels in axial, sagittal, and / or coronal image data. Alternatively, the level identification model 234 may include a model for evaluating volumetric image data and / or combined image data, such as, for example, image data from multiple imaging systems and / or combined axial, sagittal, and / or coronal image data.

[0071] The spine analysis model 252 can be a model or algorithm for analyzing anatomical features and / or performing spinal deformity assessment or diagnosis. In some embodiments, the spine analysis model 252 can associate spinal deformity assessment information (e.g., output of the spine analysis model 252) with a particular level of the spine (e.g., output of the level discrimination model 234) and / or with a particular anatomical part or structure (e.g., output of the segmentation model 232). In some embodiments, the spine analysis model 252 can include a machine learning model, such as a CNN model, an SVM model, etc. The spine analysis model 252 can be implemented by the processor 220 to perform the anatomical feature analysis 226 and / or the deformity assessment 228. In some aspects, the spine analysis model 252 can be specific to the type of spinal condition and / or spinal deformity being assessed. For example, the spine analysis model 252 can include a stenosis model 252a for stenosis assessment and / or a disc degeneration model 252b for disc degeneration assessment.

[0072] The anatomical part data 240 may include information about the anatomical parts of a patient. For example, the anatomical part data 240 may include information that identifies, characterizes, and / or quantifies different features of one or more anatomical parts, such as, for example, the location, color, shape, geometry, or other aspects of the anatomical parts. The anatomical part data 240 may provide general or patient-specific information about different anatomical parts to enable the processor 220 to perform segmentation 222, level identification 224, anatomical feature analysis 226, and / or deformation assessment 228 based on the patient image data. The image data 242 may include image data related to one or more patients and / or information about different imaging devices, such as different settings of different imaging devices (e.g., imaging device 160) and how those settings may affect images captured using those devices.

[0073] The processor 220 may be any suitable processing device configured to perform and / or execute any of the functions described herein. In some embodiments, the processor 220 may be a general-purpose processor, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processor (DSP), a dedicated graphics processing unit (GPU), or the like. In some embodiments, the processor 220 may be configured to perform one or more of the segmentation 222, the level identification 224, the anatomical feature analysis 226, and / or the deformation assessment 228. The segmentation 222, the level identification 224, the anatomical feature analysis 226, and the deformation assessment 228 may be implemented as one or more programs and / or applications coupled to hardware components (e.g., the processor 220, the memory 230, the input / output interface 250). In some embodiments, a system bus (not shown) may be configured to enable the processor 220, the memory 230, the input / output interface 250, and / or other components of the computing device 210 to communicate with each other.

[0074] Although a single processor 220 disposed on a single computing device 210 is shown in FIG. 2, it can be understood that the processor 220 may be one or more processors disposed on the same computing device and / or different computing devices. In some embodiments, the systems, devices, and methods described herein, including, for example, anatomical feature analysis and / or deformation assessment, can be implemented on a cloud platform (e.g., using one or more remote computing devices). The cloud platform can be connected to one or more databases, such as, for example, hospital databases, via a network (e.g., network 150). Thus, the systems, devices, and methods described herein can receive information (e.g., patient information, imaging data, etc.) from those databases and / or send information (e.g., spine analysis reports, visualization of the patient's anatomy, etc.) to those databases.

[0075] The input / output interface 250 can include one or more components configured to receive input and send output to other devices (e.g., imaging device 160, surgical navigation system 170, etc.). In some embodiments, the input / output interface 250 can include a user interface, which can include one or more components configured to receive input and / or present output to a user. For example, the input / output interface 250 can include a display device (e.g., a display, a touch screen, etc.), an audio device (e.g., a microphone, a speaker), a keypad, and / or other interfaces for receiving information from a user and / or presenting information to a user. In some embodiments, the input / output interface 250 can include a communication interface for communicating with other devices, which can include conventional electronics for communicating data using standard communication protocols such as Wi-Fi, Bluetooth, etc.

[0076] 2. Method Overview The systems, devices, and methods described herein can perform spinal deformity analysis (e.g., stenosis assessment, disc degeneration assessment, and / or other deformity assessment). In some embodiments, the systems, devices, and methods can perform such analysis in combination with segmentation and / or level identification. As previously described, a computing device (e.g., computing device 110, 210) for performing segmentation, level identification, and / or spinal analysis can implement one or more algorithms or models. In some embodiments, the algorithm or model can include a machine learning model that can be trained using a labeled training data set. The machine learning model can use the training data set to learn the relationship between different features in the image data and the output spinal deformity.

[0077] In some embodiments, the systems, devices, and methods described herein may perform pre-processing of image data, for example, before performing segmentation, level identification, vertebral feature analysis, and / or spinal deformity assessment. Often, image data collected using conventional imaging techniques may be of low quality. For example, a CT imaging device may be used at a low dose setting to capture images of the patient's anatomy to avoid the risk of exposing the patient to high levels of radiation. Similarly, an MRI imaging device using lower power may be used to capture images of the patient's anatomy. Such low dose or low power images may have images with a higher amount of noise. The computing devices described herein (e.g., computing devices 110, 210) may pre-process images to remove such noise, as needed, before performing segmentation, vertebral level identification, anatomical feature analysis, and / or spinal deformity assessment.

[0078] FIG. 3A shows a schematic architecture of a CNN model 350 that can be utilized to perform segmentation (e.g., both semantic and binary). The CNN model 350 can be an example of a level discrimination model, such as the segmentation model 232 described with reference to FIG. 2. The architecture of the CNN model 350 can be fully convolutional and have layer skip connections. The CNN model 350 can perform pixel-wise class assignment using an encoder-decoder architecture. The CNN model 350 can take as input raw images generated by an imaging device (e.g., the imaging device 160) and / or images generated by the imaging device after, for example, denoising, with another CNN model for denoising. The left side of the CNN model 350 is a contraction path (encoder), which includes one or more convolutional layers 360 and / or pooling layers 362. One or more images (e.g., raw or denoised images) can be presented to an input layer of CNN model 350, which through a series of convolutional layers 360 and / or pooling layers 362 can extract features from the image data. To the right of CNN model 350 is the augmentation path (decoder), which includes an upsampling or transposed convolutional layer 370 and a convolutional layer 372, resulting in an output layer 380.

[0079] In some embodiments, the CNN model 350 can be used to perform segmentation of the patient's spinal anatomy. For example, the CNN model 350 can be configured to classify portions of an image (e.g., each pixel or group of pixels) into different classes, such as bone or not, or bone, nerve, vertebral body, pedicle, process, etc. In some embodiments, a first CNN model can be configured to perform a first classification (e.g., bone or not), and the output of the first CNN model 350 can be combined and input to one or more additional CNN models 350 configured to perform one or more additional classifications (e.g., nerve or not, disc or not, disc or not, etc.). In some embodiments, the CNN model 350 can be trained to segment the patient's anatomy using a training dataset that includes images with labeled anatomical parts.

[0080] Further details of the CNN model 350 configured to perform segmentation are described in U.S. Patent Application Publication No. 2019 / 0105009, U.S. Patent Application Publication No. 2020 / 0151507, U.S. Patent Application Publication No. 2020 / 0410687, U.S. Provisional Patent Application No. 63 / 187,777, and PCT Patent Application No. PCT / US22 / 29000, which are incorporated above by reference.

[0081] FIG. 3B is a schematic diagram illustrating a CNN model 300 for performing level discrimination of a spinal anatomy, according to an embodiment. The CNN model 300 may be an example of a level discrimination model, such as the level discrimination model 234 described with reference to FIG. 2. The CNN model 300 may be configured to perform vertebral level prediction on image data including, for example, a 2D scan (e.g., DICOM slice) of an anatomical structure. In some embodiments, the CNN 300 may be configured to return one or more probability maps for each vertebral level class (e.g., thoracic, sacral, lumbar, cervical) that identify the probability that the anatomical image (or a portion of the image, e.g., a pixel or group of pixels) belongs to that particular class. The probabilities associated with the different classes may sum to one, such that the class with the greatest probability indicates the most likely class to which the image (or a portion of the image) belongs.

[0082] In some embodiments, the input to the CNN model 300 may be a contraction path (encoder) and includes multiple stacked convolution blocks 310 including one or more convolution layers and / or pooling layers. One or more images (e.g., raw images or denoised images) may be presented to the input layer of the CNN model 300, and the CNN model 300 through a series of convolution layers and / or pooling layers may extract features from the image data. The image data may include a single image (e.g., an x-ray image or a single image scan) or a set of images of a 2D scan that together form a local volumetric representation. In some embodiments, the last convolution block 310 may be directly connected to multiple dense fully connected layers 302 stacked on top of each other. The last fully connected layer 302 may be considered a network output layer that corresponds to all possible outputs. For example, the possible outputs may include all vertebra type classes (e.g., cervical, thoracic, lumbar, sacral).

[0083] As previously discussed, the CNN model 300 can be used to perform level discrimination of spinal anatomy. For example, the CNN model 300 can be configured to classify portions of an image (e.g., individual pixels or groups of pixels) into different level type classes, e.g., thoracic, sacral, lumbar, and / or cervical. In some embodiments, the CNN model 300 can be configured to classify portions of an image into different vertebral level (ordinal identifier) ​​classes, e.g., thoracic levels 1-12 (TH1-TH12), lumbar levels 1-5 (L1-L5), sacral levels 1-5 (S1-S5), and / or cervical levels 1-8 (C1-C8). In some embodiments, a first CNN model can be configured to perform a first classification (e.g., vertebral level type), and the output of the first CNN model can be combined and input to one or more additional CNN models configured to perform one or more additional classifications (e.g., ordinal identifiers). In some embodiments, the CNN model 300 can be configured to classify images by identifying a pair of spinal levels (e.g., L1 / L2, C6 / C7, etc.) or a range of spinal levels (e.g., C5-T7, L1-L4, etc.). As previously described, the CNN model 300 can be trained to identify a patient's anatomy using a training dataset that includes images with labeled anatomical parts.

[0084] Further details of the CNN model 300 configured to perform level discrimination are described in U.S. Patent Application Publication No. 2020 / 0327721 and U.S. Provisional Patent Application No. 63 / 256,306, which are incorporated above by reference.

[0085] The spine analysis model can be used to measure or determine information about one or more anatomical features and / or perform spinal deformity assessment, as described with reference to FIG. 2. In some embodiments, the spine analysis model can be implemented as a CNN model and / or other machine learning algorithm. In other words, the CNN model can be an example of a spine analysis model, such as the spine analysis model 252 described with reference to FIG. 2. In some embodiments, the CNN model can perform anatomical feature analysis in response to receiving image data, including, for example, 2D scans and / or 3D volumetric image data of an anatomical structure. In some embodiments, the CNN model can be configured to receive and / or extract information about a region of interest from the image data (e.g., parameters, characteristics, or other information of anatomical features, such as vertebrae, discs, intervertebral foramina, nerves, etc.). For example, the CNN model can be configured to receive or extract information including a shape of an anatomical feature, a surface area of ​​an anatomical feature, a distance between two adjacent anatomical features, combinations thereof, etc. The CNN model can be configured to classify one or more anatomical parts and / or structures based at least in part on an analysis of the anatomical feature information (e.g., shape, surface area, distance, etc.). In some embodiments, the CNN model can be configured to assess a condition or status of one or more anatomical parts and / or structures based on the anatomical feature information. For example, the CNN can be configured to classify one or more vertebral levels as having a mild spinal deformity, a moderate spinal deformity, a severe spinal deformity, no spinal deformity, etc. based on the anatomical feature information.

[0086] Although the segmentation model, level identification model, and spine analysis model are described separately in this section, it may be understood that one or more models for performing segmentation, level identification, and / or spine analysis may be used together, for example, in a model or algorithm that combines multiple such processes (segmentation, level identification, and / or spine analysis) together. For example, in some embodiments, a model may be configured to receive image data, perform segmentation on the image data, identify vertebrae at different levels within the image data, perform analysis of different spinal deformities, and output information regarding one or more anatomical parts and / or structures.

[0087] Further details of the training of the various models are described with reference to the flow diagrams shown in Figures 4A and 4B. Further details of the implementation of the models to perform anatomical feature analysis and / or spinal deformity assessment are described with reference to the flow diagrams shown in Figures 5-7B. The methods illustrated in these flow diagrams may be performed by one or more devices such as those described with reference to Figures 1 and 2, including, for example, computing devices 110, 210.

[0088] 3. Training the Model for Spine Analysis In some embodiments, the systems, devices, and methods described herein can perform spinal analysis using a model trained to analyze and / or diagnose a condition or deformity associated with a patient's spine. For example, the systems, devices, and methods described herein can use a model trained to extract one or more features from anatomical image data and assess spinal deformity based on the extracted features. Examples of spinal deformity can include spinal stenosis, spinal degeneration analysis, vertebral fractures, spondylolisthesis, scoliosis, tumors, and / or herniated discs.

[0089] FIG. 4A is a flow chart of a method 400 for training a spine analysis model (e.g., spine analysis model 252 of FIG. 2), according to some embodiments. The spine analysis model can be used to perform spinal deformity assessment and / or spinal deformity diagnosis. The method 400 can include, at 410, reading image data and / or other input data from a training dataset. The training dataset can include input images of anatomical structures (e.g., vertebrae, nerves, discs, etc.) and corresponding output images of the anatomical structures with the anatomical structures labeled. In some embodiments, the training dataset can optionally include other input data including, for example, segmentation data identifying different anatomical structures in the image and / or level identification data identifying different levels of the spine. In some embodiments, the input image can be merged or combined with the segmentation data and / or level identification data to provide a combined image including image data indicative of tissue appearance and its classification (e.g., assignment of anatomical parts and / or structures based on the segmentation data and / or assignment of levels based on the level identification data). Alternatively, the image data, segmentation data, and / or level discrimination data may be provided separately as inputs to the model.

[0090] The output image may include labels associated with different output classifications. In some embodiments, the output image may include labels that provide information regarding the status or condition of one or more anatomical parts of the structure. For example, the output image may include a label that identifies whether a particular vertebral level or range of vertebral levels has a spinal deformity. Alternatively or additionally, the output image may include a label that identifies a type of spinal deformity and / or a degree or grade of the deformity (e.g., Grade I-Grade IV, or none, mild, or severe). In some embodiments, the output image may include a label that identifies one or more anatomical features of interest (e.g., anatomical features relevant to spinal deformity assessment), including one or more anatomical parts and / or parts of the structure, or visual or geometric features that characterize one or more anatomical parts and / or parts of the structure (e.g., distance between two anatomical parts; length, width, circumference, or other geometric parameters of an anatomical part; curvature of the spine or part thereof). In some embodiments, the output image may include a label that identifies a vertebral level type and / or vertebral level associated with a spinal condition and / or deformity. In an exemplary implementation, the output images may include one or more labels indicating vertebral levels, information corresponding to anatomical feature analysis (e.g., shape, surface area, distance, etc.), and / or information corresponding to spinal deformity assessment. In some embodiments, the images (e.g., input images in a training dataset and / or output images) may be 2D images of the anatomical structure showing one or more views, such as an axial view, a sagittal view, a coronal view, etc., of the anatomical structure. The images may be grouped into multiple batches to train the spine analysis model. Each image in a batch may include images representing a 2D region of the anatomical structure and / or a series of slices of a 3D volume. A computing device (e.g., computing device 110, 210) may read the image data by loading one or more batches of images for further processing.

[0091] Optionally, at 420, several 2D images of the anatomical structure may be combined to generate a 3D volume of the anatomical structure. For example, as described above, the 2D images may depict a particular view of the anatomical structure, such as an axial view, a sagittal view, a coronal view, etc. The 2D images of different views of the anatomical structure may be combined to generate a 3D volume of the anatomical structure. For example, an image representing an axial view of the anatomical structure may be combined with an image representing a sagittal view of the anatomical structure to generate a 3D volume of the anatomical structure. The 3D volumetric image data may then be used to train a model.

[0092] In some embodiments, image data (e.g., 2D image data, 3D image data, etc.) may be augmented. Data augmentation may be performed on image data to create a more diverse set of images. Each input image and its corresponding output image may undergo the same data augmentation, and the resulting input and output images may be stored as new images in the training dataset. Data augmentation may include applying one or more transformations or other data processing techniques to the images. These transformations or processing techniques may include rotation, scaling, translation, horizontal flip, Gaussian and / or Poisson distributed additive noise, Gaussian blur, and the like. Data augmentation may be performed on any image type, including, for example, X-rays, CT scans, and / or MRI scans, and any image view (e.g., axial, sagittal, coronal).

[0093] Optionally, at 430, a multidimensional region of interest (ROI) may be selected and / or defined within the image data. In some embodiments, the ROI may be defined based on a predefined parameter (e.g., size of region or multidimensional stride). In some embodiments, overlapping ROIs may be defined, while in other embodiments, non-overlapping ROIs may be defined. In some embodiments, the predefined parameter may be adjusted based on a size or type of image data and / or a type of spinal deformity assessment (e.g., stenosis assessment, disc degeneration assessment, other deformity assessment, etc.). In some embodiments, image data from a 3D scan volume may be combined with segmentation data, level identification data, other image data (e.g., CT or X-ray combined with MRI), and / or other data to provide input data with higher dimensionality. For example, 3D volumetric image data (e.g., 3D volumes generated from 2D images and / or 3D information from a medical imaging source such as imaging device 160) may be combined with other 3D information (e.g., manual and / or autonomous segmentation data, image data from another imaging source) to generate higher dimensional image data or ROI. In an exemplary embodiment, each ROI or 3D region of image data may include (1) information regarding voxel distribution along multidimensional axes (e.g., X, Y, and Z axes), (2) appearance information of the anatomical part captured by the imaging system for each voxel, and (3) segmentation data for each voxel indicating the classification of the anatomical part.

[0094] In some embodiments, the image data may be processed by one or more trained segmentation models (e.g., the segmentation model of FIG. 3A) to segment anatomical structures of interest in the image data. For example, a computing device may be configured to input the image data into the trained segmentation model, such that the segmentation model generates an output that classifies or labels one or more anatomical parts or structures of interest. This segmentation output may then be used along with the image data to train a spine analysis model (and optionally combined with the image data to generate multi-dimensional data as described above).

[0095] In some embodiments, the image data may be processed by one or more level discrimination models (e.g., the level discrimination model of FIG. 3B) to identify one or more vertebral levels of interest. For example, the computing device may be configured to input the image data into a trained level discrimination model such that the level discrimination model generates an output that classifies or labels a vertebra or range of vertebrae within the image data or ROI. This level discrimination output may then be used along with the image data to train a spine analysis model (and optionally combined with the image data to generate multi-dimensional data as described above).

[0096] At 440, the spine analysis model may be trained using the training dataset. In some embodiments, the spine analysis model may be trained using image data including original image data and / or augmented image data, as well as other input data (e.g., segmentation data, level discrimination data, etc.). In some embodiments, the training may be supervised. The training may include inputting input images into the spine analysis model and minimizing the difference between the output of the spine analysis model and an output image (i.e., an image with an associated label) in the training dataset that corresponds to the input image. In some embodiments, the spine analysis model may be a CNN model, whereby one or more weights of the function may be adjusted to better approximate the relationship between the input image and the output image. Further details of training the CNN model are described with reference to FIG. 4B. In some embodiments, the training may be unsupervised. For example, the spine analysis model may rely on distances between feature vectors to classify known data points.

[0097] The validation dataset can be used to evaluate one or more performance metrics of the trained spine analysis model. Similar to the training dataset, the validation dataset can include input images of anatomical structures (e.g., vertebrae, nerves, discs, etc.) and output images including labels associated with the anatomical structures and / or assessments of spine deformity. The validation dataset can be used to check whether the trained spine analysis model meets a particular performance metric or whether further training of the spine analysis model may be required. At 450, the input images of the validation dataset can be run through the trained spine analysis model to obtain an output. At 460, based on the output of processing the validation dataset, one or more performance metrics can be calculated. For example, the output of the validation dataset can be compared to an output image corresponding to the input image, and the difference between the model's output and the output image can be evaluated on a qualitative and / or quantitative scale. Based on the difference between the model's output and the output image corresponding to the input image, different performance metrics can be calculated. For example, the number or percentage of correctly or incorrectly classified pixels (or groups of pixels) can be determined and / or the Sorensen-Dice coefficient can be calculated.

[0098] At 470, the computing device can determine whether training is complete (e.g., performance of the trained spine analysis model is sufficient and / or a certain number of training iterations have been met) or whether further training is required. In some embodiments, the computing device can continue to periodically repeat the training iterations (i.e., return to 410-460) until the performance of the trained model is no longer improving by a predetermined amount (i.e., the performance metrics of the later training iterations 410-460 do not differ from the performance metrics of the previous training iterations 410-460 by a predetermined threshold or percentage). If the model is not improving, the spine analysis model may be overfitting the training data. In some embodiments, the computing device can continue to periodically repeat the training iterations (i.e., return to 410-460) until the performance metrics of the training iterations 410-460 reach a certain predetermined threshold indicating sufficient performance. In some embodiments, the computing device may continue to periodically repeat the training iterations (i.e., returning to 410-460) until the predetermined number of iterations has been met (i.e., the spine analysis model has been trained the predetermined number of times).

[0099] Once the spine analysis model is sufficiently trained (470: YES), the spine analysis model may be stored, for example, in a memory (e.g., memory 230) at 480. The stored spine analysis model may be used by the computing device in an inference or prediction process, for example, to perform spinal deformity assessment and / or spinal deformity diagnosis on new image data of the patient.

[0100] 4B is a flow chart providing a more detailed description of training 440 a spine analysis model implemented as a neural network, such as a CNN, according to an embodiment. The neural network model can be trained by tuning or adjusting parameters of the neural network model such that different portions of image data can be classified based on features extracted from those portions. Once trained, the neural network can be used to perform spinal deformity assessment and / or spinal deformity diagnosis (e.g., stenosis assessment, disc degeneration assessment, other deformity assessment, etc.) on multiple images (e.g., 2D scans of a patient's anatomy and / or 3D volumes generated from the 2D scans).

[0101] The method 440 may include, at 441, inputting a batch of image data from a training dataset into a neural network. As previously described, the training dataset may include input images of a patient's anatomy and corresponding output images of labeled patient's anatomy (e.g., anatomical components such as vertebrae that are labeled with information related to anatomical feature analysis and / or spinal deformity assessment). The batch of images may be read one at a time from the training dataset and processed using the neural network. In some embodiments, the batch of images may include different views of the anatomy (e.g., axial, sagittal, etc.). The different images may be combined to generate a 3D volume of the anatomy.

[0102] The batch of images may be passed through the layers of the neural network in a standard forward pass at 442. The forward pass may return an output or result that may be used to calculate a value of a loss function at 444. The loss function or objective function represents a function used to evaluate the difference between a desired output (as reflected in an output image corresponding to the input image) and the output of the neural network. The value of the loss function may indicate a measure of that difference between the desired output and the output of the neural network. In some embodiments, the difference may be represented using a similarity metric including, for example, mean squared error, mean average error, or categorical cross entropy. The value of the loss function may be used to calculate an error gradient, which may be used to update one or more weights or parameters of the neural network at 446. The weights and parameters may be updated to reduce the value of the loss function during subsequent passes through the neural network.

[0103] At 448, the computing device can determine whether the training has cycled through the complete training data set, i.e., whether an epoch is complete. If the epoch is complete, the process can proceed to 450, where the validation data set is used to evaluate performance metrics of the trained spine analysis model. Otherwise, the process can return to 441, where the next batch of images is passed to the neural network.

[0104] Although not described in connection with Figures 4A and 4B, the method 400 can also be used to train segmentation and / or level discrimination models. Detailed examples of such training are described in U.S. Patent Application Publication No. 2019 / 0105009, U.S. Patent Application Publication No. 2020 / 0151507, U.S. Patent Application Publication No. 2020 / 0410687, U.S. Provisional Patent Application No. 63 / 187,777, PCT Patent Application No. PCT / US22 / 29000, U.S. Patent Application Publication No. 2020 / 0327721, and U.S. Provisional Patent Application No. 63 / 256,306, all of which are incorporated by reference above.

[0105] 4. Spine analysis 5 is a flowchart of a method 500 for performing spinal analysis, according to some embodiments. Method 500 may be performed by a computing device, such as computing device 110, 210. In some embodiments, the systems, devices, and methods described herein may use one or more spinal analysis models, such as trained per method 400 described in connection with FIGS. 4A and 4B. Alternatively or additionally, in some embodiments, the systems, devices, and methods described herein may analyze certain anatomical features and assess one or more spinal conditions or deformities based on such analysis.

[0106] The method 500 may include, at 510, reading a batch of images from the patient image data. The images may be new images acquired of the patient's anatomical structure of interest. The images may be, for example, 2D scans of a 3D volume of one or more anatomical structures. In some embodiments, the images may include CT images, MRI images, and / or X-ray images. In some embodiments, the images may include axial, sagittal, and / or coronal views. In some embodiments, the images may be pre-processed. For example, one or more images may be denoised using a model for denoising image data. Alternatively or additionally, the images may be processed using other techniques, such as, for example, filtering, smoothing, cropping, normalising, resizing, etc. In some embodiments, an estimated time warp may be applied to one or more images, with a predetermined number of warped images being created for each input image. These warped images may produce inference results that are robust with respect to small variations in brightness, shrinkage, orientation, etc.

[0107] At 520, segmentation may be performed on the patient image data. For example, the patient image data may be processed by one or more segmentation models (e.g., trained segmentation model 232 of FIG. 2 or CNN 350 of FIG. 3A). In some embodiments, the patient image data may be segmented based on predefined and / or selected ROIs (e.g., ROIs selected in step 430 of FIG. 4A, etc.). For example, the one or more segmentation models may process the ROIs (e.g., selected ROIs and / or predefined ROIs) to segment the patient image data and define or indicate the size and shape of the ROIs within an anatomical structure. The segmentation models may be trained to identify one or more anatomical parts of interest from the image data. For example, the trained segmentation models may be used to identify bony structures (e.g., vertebral bodies, pedicles, transverse processes, laminas, and / or spinous processes) and / or one or more soft tissue structures of interest (e.g., nerves, intervertebral discs, etc.). Suitable examples of performing segmentation are described in U.S. Patent Application Publication No. 2019 / 0105009, U.S. Patent Application Publication No. 2020 / 0151507, U.S. Patent Application Publication No. 2020 / 0410687, U.S. Provisional Patent Application No. 63 / 187,777, and PCT Patent Application No. PCT / US22 / 29000, which are incorporated by reference above.

[0108] The image data may be processed 530 using a level identification model (e.g., level identification model 234). Some exemplary level identification models include a vertebra-based level identification process, a disc-based level identification process, an axial image-based level identification process, a sagittal or coronal image-based level identification process, etc. Further details of such models are described in U.S. Provisional Patent Application No. 63 / 256,306, incorporated above by reference. For example, the systems, devices, and methods may assign a level type (i.e., cervical (C), thoracic (TH), lumbar (L), and / or sacral (S)) or vertebral level (e.g., C1-S5, or C1-C7, TH1-TH12, L1-L5 (L6), and / or S1-S5) to an image or portion of an image based on morphological and spatial relationships determined using vertebral-based level identification and / or disc-based level identification, a prediction of vertebral level type and / or vertebral level determined using axial image-based level identification, and / or a prediction of vertebral level or vertebral level range determined using sagittal or coronal image-based level identification. In some embodiments, a vertebral type may be assigned to one or more sub-volumes or groups of vertebrae (e.g., using a level identification model), and then a vertebral level or ordinal identifier may be assigned to one or more sub-volumes or groups of vertebrae based on the morphological and spatial relationships between the vertebrae, the orientation of the patient's anatomy within the image data, the overall distribution of level types, etc. In some embodiments, an index can be assigned and counts can be used to assign an ordinal identifier. For example, counting for the lumbar vertebrae can start at L5 (or L6) if the sacrum is included in the image data, and start at L1 if the thoracic vertebrae are included in the image data. Similar counts can be used for each of the other vertebrae (e.g., cervical, sacral, and thoracic). Ordinal identifiers can be assigned to each group of anatomical components belonging to a level type (e.g., C, TH, L, S) based on the distribution of anatomy and all other levels.

[0109] Alternatively or additionally, in some embodiments, a level discrimination model trained to assign vertebral levels to one or more vertebrae can be used to assign vertebral level or ordinal identifiers (e.g., C1-S5, or C1-C7, TH1-TH12, L1-L5 (L6), and / or S1-S5) to one or more portions of image data. For example, the level discrimination model can be trained using input images of one or more vertebrae and corresponding output images including labels identifying the vertebral levels of one or more vertebrae. Further details of the use of a trained level discrimination model are described with reference to FIG. 6A.

[0110] The patient image data may be further processed through one or more spinal analysis models to perform spinal deformity assessment at 540. In some embodiments, the image data may be combined with the segmentation data and / or level identification data before being processed by the spinal analysis models. In some embodiments, the image data, segmentation data, and / or level identification data may be provided separately as inputs to the spinal analysis models to assess spinal deformity.

[0111] At 550, the image data (or a portion of the image data) can be input to a stenosis assessment process. For example, the patient image data (or a portion of the image data) can be processed to analyze anatomical features such as the spinal cord, nerve roots, the dural sac, etc. In particular, the stenosis assessment can be performed based on the analysis of anatomical features such as determining the surface area of ​​the dural sac and / or the area of ​​the spinal cord, or tracing the nerve roots. Further details of an exemplary stenosis assessment process are described with reference to FIG. 7A. At 560, the image data (or a portion of the image data) can be input to a disc degeneration assessment process. For example, the patient image data (or a portion of the image data) can be processed to analyze anatomical features such as the intervertebral disc. In particular, the disc degeneration assessment can be performed based on the analysis of anatomical features such as the strength, shape and / or surface area of ​​the disc, or the analysis of the distance between two adjacent vertebrae, etc. Further details of an exemplary disc degeneration assessment process are described with reference to FIG. 7B. At 570, the image data (or a portion of the image data) may be input to other suitable deformity assessment processes (e.g., other deformities such as herniated discs, osteophyte formation, spondylolisthesis, etc.) to identify other deformities. For example, the patient image data (or a portion of the image data) may be processed to analyze any suitable anatomical features (e.g., vertebral bodies, pedicles, laminae, nerves, discs, etc.) to assess other deformities (e.g., as shown with reference to FIG. 11 ).

[0112] At 580, a spinal deformity assessment can be presented to a user (e.g., a surgeon). The spinal deformity assessment can include a type of deformity (e.g., stenosis, disc degeneration, and / or other deformity) and / or a severity or grade of the deformity (e.g., mild, moderate, severe, or grade I-grade IV, etc.). In some embodiments, the presentation of the spinal deformity assessment can include a vertebral level associated with the spinal deformity assessment, for example based on the output of the level identification at 530, along with information indicative of the spinal deformity assessment. As an example, an assessment "L5-S1: Mild Spinal Stenosis" as shown in FIG. 9A can be presented to the user. In particular, the assessment includes an ordinal identifier L5-S1 indicating that the anatomical portion in the patient image data includes the lumbar vertebrae starting at L5 to the sacrum S1. The assessment also includes an output from the stenosis assessment process 550 indicating mild spinal stenosis in the patient. In some embodiments, when spinal deformity assessment is performed on multiple sets of patient image data (e.g., image data having different views, such as axial and sagittal views, image data captured in different imaging sessions, image data captured using different imaging devices), assessments on the different sets of patient image data can be merged together or presented together. For example, when different sets produce different spinal deformity assessments (e.g., one assessment is mild and another is severe), the different assessments can be merged (e.g., averaged with or without weighting factors, majority voting, etc.), for example according to a predefined scheme, or presented together to a user (e.g., surgeon). As another example, when stenosis assessment is performed on axial and sagittal image views of a patient, assessments from both of these views can be merged and presented to a user, or presented together to a user. In some embodiments, the output from processing the different views can complement each other. For example, the output from the analysis of the axial view can include a spinal deformity assessment that can be confirmed with the assessment obtained from the analysis of the sagittal view images. Further examples of presenting spinal deformity assessments are described below with reference to Figures 8-10C.

[0113] Optionally, one or more virtual representations of the patient's anatomy may be generated at 582, for example, for visualization in pre-operative planning (e.g., via computing device 110, 210) and / or image-guided surgery (e.g., via surgical navigation system 170). In some embodiments, a 3D anatomical model may be generated based on the image data that may be used to generate the virtual representation of the patient's anatomy for visualization. In some embodiments, the 3D anatomical model may be converted or used to generate a polygonal mesh representation of the patient's anatomy (or a portion thereof). Parameters of the virtual representation (e.g., volume and / or mesh representation) may be adjusted with respect to color, opacity, mesh thinning, etc. to provide a user (e.g., surgeon) with different views of the patient's anatomy. In some embodiments, the virtual representation may be a 2D view, e.g., a sagittal or coronal view, of the patient's anatomy, with information regarding the type or level of the labeled vertebral levels and / or an analysis of the anatomical features (e.g., surface area of ​​the spinal cord or dural sac, shape of the intervertebral disc, etc.). In some embodiments, the virtual display may be a 3D view, such as, for example, a 3D structure or model of the patient's spine and / or adjacent anatomical structures (e.g., nerves, discs, etc.). In some embodiments, the anatomical model of the patient's anatomy may be used to provide a virtual or augmented reality image for display by a computer-assisted surgery system, such as, for example, surgical navigation system 170. In such a system, a virtual 2D or 3D view of the patient's anatomy may be displayed over an actual portion of the patient's anatomy (e.g., the surgical site).

[0114] At 590, the output of the spinal deformity assessment and / or the spinal analysis model may be stored in a memory (eg, memory 230).

[0115] i. Anatomical segmentation As previously described in connection with 520 of FIG. 5, the systems, devices, and methods described herein can be configured to perform autonomous spine anatomical segmentation. In some embodiments, the segmentation can be performed using pre-trained models, including those described in U.S. Patent Application Publication No. 2019 / 0105009, U.S. Patent Application Publication No. 2020 / 0151507, U.S. Patent Application Publication No. 2020 / 0410687, U.S. Provisional Patent Application No. 63 / 187,777, and PCT Patent Application No. PCT / US22 / 29000, all of which are incorporated by reference above. For example, the segmentation can be performed using a pre-trained modified 2DU-Net neural network.

[0116] In some embodiments, segmentation can be performed on MRI images, including, for example, T2-weighted scans. The systems and devices described herein (e.g., computing devices 110, 210) can be configured to load patient image data and perform automatic selection of a T2-weighted sequence that provides better visualization of one or more anatomical parts of interest (e.g., dural sac, nerve roots, intervertebral disc, etc.). T2 hyperintensity can be associated with fluid mass tissue, and T2 hypointensity can reflect hypercellularity. Segmentation can be performed in multiple planes, such as, for example, sagittal and axial planes. In some embodiments, a first model (e.g., a neural network) can be trained and used to perform segmentation of axial images, and a second model (e.g., a neural network) can be trained and used to perform segmentation of sagittal images. Alternatively, a single model can be trained and used to perform segmentation of images in multiple planes (e.g., axial and sagittal).

[0117] In some embodiments, pre-processing of the imaging data may be performed prior to processing the imaging data by the segmentation model to segment the imaging data, For example, such pre-processing may include one or more algorithms or models that analyze the quality, rotation, sequence, and / or other like parameters of the imaging data to ensure that the imaging data is accurate and ready for input into the segmentation model.

[0118] The segmentation model can be configured to segment the imaging data into a plurality of classes, for example, two to about twenty classes, including sixteen classes. For example, when used to segment MRI image data, the segmentation model can be configured to segment the imaging data into three or more classes and up to about twenty classes, including sixteen classes. In some embodiments, the segmentation model can be configured to segment the MRI imaging data into classes corresponding to one or more anatomical parts of interest without segmenting surrounding anatomical parts of no interest. For example, when assessing spinal canal stenosis and / or disc degeneration, the segmentation model can be trained and used to segment the MRI imaging data into two or more classes, including one or more of the dural sac, annulus fibrosus, and nucleus. Alternatively, when assessing stenosis, the segmentation model can be trained and used to segment the MRI imaging data into (1) two classes, for example, whether or not it is a dural sac, or (2) three classes, for example, whether it is a dural sac, nerve root, not a dural sac, or not a nerve root. Alternatively, when assessing disc degeneration, a segmentation model can be trained and used to segment MRI imaging data into three classes, such as annulus fibrosus, nucleus, and all other anatomical structures. Limiting the segmentation to a particular anatomical portion of interest can reduce computational requirements (e.g., by having a smaller neural network), provide easier and / or faster validation, and require less training and / or manual marking.

[0119] ii. Level Identification 6A is a flow chart of an exemplary level identification process 530, according to some embodiments. As previously described, one or more level identification models may be used to identify vertebral levels of different vertebrae depicted in image data.

[0120] At 531, an ROI may be selected within the image data. The ROI may be an entire volume of the image data or a portion of the image data (e.g., a sub-volume, one or more 2D images). The ROI may be selected to include a particular anatomical component or structure, such as one or more vertebrae, discs, nerves, etc. Alternatively, the ROI may be selected to encompass a particular region, volume, and / or subgroup of pixels within the image. In some embodiments, the ROI may be selected by a user (e.g., a surgeon). For example, a surgeon may wish to evaluate one or more particular anatomical components or structures, such as one or more vertebrae, discs, nerves, etc., for deformity and may select an ROI to identify the particular anatomical component or structure within the image data. In some embodiments, the ROI may be selected autonomously (e.g., by a computing device) based on the type of image data, the size of the image data, and / or the type of spinal deformity assessment.

[0121] At 532, the method may include selecting a 2D image associated with the ROI. For example, consider that the ROI selected at 531 identifies a volume of interest within the anatomy. At 532, 2D images of that volume may be selected. These images may be a 2D axial view of the volume of interest, a 2D sagittal view of the volume of interest, a 2D coronal view of the volume of interest, and / or combinations thereof. In certain embodiments, consider that the ROI selected at 531 includes an anatomical portion of interest, such as a particular vertebra. At 532, 2D images of that vertebra may be selected. These images may be a 2D axial view of the vertebra, a 2D sagittal view of the vertebra, a 2D coronal view of the vertebra, and / or combinations thereof.

[0122] Optionally, these 2D images of the ROI may be combined into a 3D ROI at 533. For example, if the ROI identifies a particular vertebra and 2D axial, 2D sagittal, and / or 2D coronal views of that vertebra are selected at 532, these axial, sagittal, and / or coronal images may be combined to generate a 3D volume of that vertebra at 533. In some embodiments, the 2D images may be used to generate a 3D model or representation of one or more anatomical parts within the ROI.

[0123] At 534, for each selected ROI (e.g., 2D images associated with the ROI and / or 3DROI), image data associated with that ROI can be processed with one or more level discrimination models (e.g., the level discrimination CNN 300 of FIG. 3B). The image data can include axial, sagittal, and / or coronal image scans including one or more vertebrae. The level discrimination models can be trained to predict the vertebral level of each vertebra based on the axial, sagittal, and / or coronal image scans associated with the ROI. In some embodiments, multiple scans associated with each vertebra in the ROI can be processed separately to provide a respective level discrimination prediction for that vertebra. Alternatively, multiple scans associated with each vertebra (e.g., axial, sagittal, and / or coronal) can be processed together to provide a single level discrimination prediction for that vertebra (or can be processed in a batch that each provides a level discrimination prediction for that vertebra).

[0124] The level discrimination model can be trained to generate a probability map representing the probability of assigning an anatomical component (e.g., each vertebra) to a class (e.g., a vertebral level or order identifier). In other words, the output of the level discrimination model can include a probability map for each class (e.g., each vertebral level) for each anatomical component. For example, the output of the level discrimination model can include a per-class probability for each pixel (or group of pixels) of the image data. More specifically, the level discrimination model can be configured to classify the image data into one of a plurality of classes (e.g., vertebral levels). Thus, the level discrimination model can be configured to generate, for each pixel or group of pixels in the image, a probability that the pixel or group of pixels belongs to any one of the classes from a plurality of classes. The multiple classes can correspond to a plurality of vertebral levels or order identifiers (e.g., TH1-TH12, L1-L5, S1-S5, and / or C1-C7).

[0125] If more image data is associated with the selected ROI (535:NO), process 530 may return to 532 to repeat the process with the additional image data (e.g., one or more additional 2D scans or 3D sub-volumes). Otherwise, process 530 proceeds to assign a level identifier or range of level identifiers for the ROI at 536.

[0126] At 536, based on the output of the level discrimination model (e.g., the probability map), a vertebral level or order identifier (e.g., C1-S5, or C1-C7, TH1-TH12, L1-L5 (L6), and / or S1-S5) can be assigned to one or more vertebrae in the ROI. For example, one or more vertebrae can be associated with different portions of the image data (e.g., different portions of a 2D axial, sagittal, and / or coronal scan, or different portions of a 3D sub-volume). When processing the image data with the level segmentation model, the model can return an output that can assign a particular class (e.g., vertebral level) to the portion of the image data corresponding to each vertebra. In some embodiments, multiple images (e.g., 2D axial, sagittal, and / or coronal images) can be associated with a particular vertebra, and the level discrimination model can predict different vertebral levels for the different images. For example, a set of axial images associated with a selected vertebra may include 80% assigned a first class (e.g., L1) and 20% assigned a second class (e.g., L2). In some embodiments, the class for a selected vertebra may be selected to be the class with the greatest number of axial images assigned to it. Thus, for a set of axial images of a vertebra where 80% are labeled "L1" and 20% are labeled "L2", the vertebra may be assigned the level "L1". Alternatively, other criteria (e.g., a predetermined number or percentage of axial images associated with a class) may be used to determine the class (vertebral level) assigned to a vertebra.

[0127] If all of the ROIs have not been processed (537: NO), process 530 can return to 531 and another ROI can be selected. At 538, the level identification data (e.g., vertebral level or ordinal identifier) ​​assigned to the ROI can be stored in memory (e.g., memory 230).

[0128] Although an exemplary method for performing level identification is described in connection with FIG. 6A, it should be readily understood that other suitable methods can be used for level identification, including the methods described in U.S. Provisional Patent Application No. 63 / 256,306, incorporated by reference above, and the disc level identification process described below.

[0129] In some embodiments, the systems, devices, and methods described herein can perform level identification of the intervertebral disc. For example, FIG. 12 illustrates an exemplary process 1130 for performing autonomous intervertebral level identification, according to an embodiment. The level identification process 1130 can be particularly suitable for grouping and assigning levels to MRI axial scans. In MRI imaging, axial scans can be performed continuously, as in CT, but can also be performed in groups or blocks around the intervertebral disc. Typically, each group of MRI axial images around the intervertebral disc includes at least some images, such as a set number of images, e.g., about five or more images, where the set number of images represents a scan volume or axial volume. Each of the axial volumes has a different scan angle that is parallel or substantially parallel to the intervertebral disc angle. In practice, most axial scan volumes have problems related to, for example, the volume itself, the number of images in the volume, etc. Therefore, in order to properly separate and identify the intervertebral discs, it can be beneficial to start with a sagittal image analysis, where the intervertebral discs are more easily identified and differentiated from each other.

[0130] At 1131, one or more sagittal images (e.g., one sagittal image or sagittal volume) of a portion or part of the spine can be selected. At 1132, segmentation results or segmentation data of the sagittal images can be analyzed to identify and separate the intervertebral discs. As mentioned above, the segmentation data can be generated, for example, by a segmentation model trained to segment the intervertebral discs starting from a particular intervertebral level. After separating the intervertebral discs in the sagittal images, at 1133, level identifiers can be assigned to the intervertebral discs, for example, by counting from the lower level and assigning a level identifier to each level.

[0131] The axial scans allow the location of the discs (e.g., real-world location) to be mapped to the axial images or volumes. The axial images may be pre-separated into axial volumes (i.e., a set of adjacent axial scans) based on their location and angle. Although the location and angle of the axial scans may allow some degree of separation between different volumes, problems arise when a volume is inadvertently scanned, for example, with an additional image in the volume or an image where two levels or discs are connected. Thus, at 1135, the location of the discs obtained from the sagittal images may be mapped to the axial volumes to improve separation of the axial volumes. In other words, the separated axial volumes may be analyzed using information obtained from the segmentation and level identification of the sagittal images such that the real-world location of the discs may be determined. Such analysis may allow the determination of which axial volume corresponds to which disc and whether there is a problem with the axial volume. For example, if it is determined at 1136a that there are two discs from the sagittal images located within a single axial volume, then at 1137a the axial volume can be separated into two portions or volumes, each portion or volume relating to a separate disc. Alternatively or additionally, if it is determined at 1136b that the axial volume is overscanned (e.g., has more than a set number of images), then at 1137b a subset of the axial images within that volume, such as a set number of images around or closest to the disc, can be selected as the volume, and the other images are discarded (i.e., not used in the spinal deformity analysis). After performing such analysis and cleaning, at 1138 the cleaned axial volumes can be bundled or associated to sagittal levels, for example, to have visualization, separation, and identification of each level in the axial and sagittal planes.

[0132] FIG. 13A visually illustrates a flow for identifying different MRI image sets and assigning levels to the different MRI image sets, according to an embodiment. The flow of FIG. 13A can be similar to the segmentation and level identification process described above. The process can start with a sagittal image or images at 1202, and segmentation can be performed in the sagittal plane. Based on the segmentation data, the intervertebral discs can be isolated and identified, as shown at 1204. Axial volumes, shown as light bands at 1206, can be registered to each of the levels in the sagittal image. In some embodiments, the axial volumes can be cleaned as described in connection with the level identification process 1130. Then, at 1208, each axial volume can be assigned to a different level.

[0133] 13B provides a visualization of a set of axial volumes 1220a-1220f combined with a sagittal image or volume 1210, according to an embodiment. In some embodiments, such visualization or similar visualizations can be provided, for example, to a surgeon to facilitate surgical planning and / or surgery. For example, visualization of sagittal and axial image data can be provided on a surgical navigation system (e.g., surgical navigation system 170) or other computing device.

[0134] iii. Spinal deformity assessment 6B is a flow chart of an exemplary spinal deformity assessment process 540, according to some embodiments. The spinal deformity assessment process 540 may be generally applicable to the analysis of a variety of different spinal deformities, including, for example, spinal stenosis, disc degeneration, vertebral fractures, etc. At 541, an ROI may be selected in the image data (e.g., similar to 531 in FIG. 6A). The ROI may be the entire volume of the image data or a portion of the image data (e.g., a sub-volume, one or more 2D images). The ROI may include one or more anatomical structures or portions thereof.

[0135] At 542, 2D images associated with the ROI may be selected (e.g., similar to 532 in FIG. 6A). These images may include an axial view of the patient's anatomy, a sagittal view of the patient's anatomy, a coronal view of the patient's anatomy, and / or combinations thereof. Optionally, at 543, these 2D images may be combined to generate a 3D ROI (e.g., similar to 533 in FIG. 6A).

[0136] At 544, one or more anatomical features of the patient's anatomy can be analyzed in the selected image data (e.g., 2D image data, or 3D volume). For example, the selected image data can include one or more anatomical parts of interest, such as the spinal cord, the dural sac, an intervertebral disc, a nerve root, etc. The selected image data can be analyzed to evaluate and / or determine one or more anatomical features or characteristics associated with the anatomical parts of interest. These features or characteristics can include the shape, surface area, volume, and / or distance between two or more anatomical parts of the anatomical parts, etc. For example, the surface area or portion of the spinal cord, the surface area or portion of the dural sac, the nerve roots (e.g., lateral and intracameral nerve roots), the shape and / or surface area of ​​the intervertebral disc, the distance between two adjacent vertebrae, etc. can be determined and / or analyzed in the image data to evaluate different types of deformation.

[0137] If more image data is associated with the selected ROI (545:NO), process 540 returns to 542 so that the process can be repeated with the additional image data (e.g., one or more additional 2D scans or 3D sub-volumes). Otherwise, process 540 proceeds to 546.

[0138] At 546, the computing device may optionally retrieve past spine analysis data for the patient. This past spine analysis data may include data related to previous evaluations of the patient's previously collected image data. For example, patient image data collected at a previous time (e.g., one or more years, months, days, etc.) may have been evaluated by a physician and / or processed using the spinal deformity assessment process 540. The physician's evaluation and / or output of the spinal deformity assessment process 540 of the previous image data may be stored as past spine analysis data. The past spine analysis data may then be retrieved in a later spine analysis process 540 to inform, for example, changes and / or rates of change of one or more features or characteristics of one or more anatomical parts over time. For example, the past spine analysis data may include information regarding rates of change of the spinal cord surface area, the surface area of ​​the dural sac, the shape and / or surface area of ​​the intervertebral disc, the distance between two adjacent vertebrae, etc.

[0139] At 547, the image data (e.g., 2D images associated with the ROIs and / or 3DROIs) may be processed by one or more spine analysis models to assess spinal deformity (e.g., stenosis, disc degeneration, disc herniation, etc.) of one or more anatomical portions and / or structures within the image data. For example, the image data may be input to a stenosis assessment process, as described in more detail herein with respect to FIG. 7A. Similarly, the image data may be input to a disc degeneration assessment process, as described in more detail herein with respect to FIG. 7B. As previously mentioned, in some embodiments, the image data may be combined with segmentation data and / or level identification data before being processed by the spine analysis models. In some embodiments, the image data, segmentation data, and / or level identification data may be provided separately as input to a spine analysis model for assessing spinal deformity. The spine analysis model may process the image data (optionally with the segmentation and / or level identification data) to predict or determine the presence or absence of deformity. Additionally or alternatively, the spine analysis model can process the image data (optionally with segmentation and / or level discrimination data) to grade the severity of the deformity (e.g., no deformity, mild deformity, moderate deformity, severe deformity, etc.).

[0140] In some embodiments, the spine analysis model can be a machine learning model (e.g., CNN) that can predict or determine the status or condition of the anatomical structures and / or portions thereof in the image data based on the segmentation data generated at 520, the level identification data generated at 530, the anatomical feature analysis performed at 544, and / or other features extracted from the image data, and / or the past spine analysis data retrieved at 546. In some embodiments, the spine analysis model can apply one or more criteria to evaluate the status or condition. For example, a threshold and / or a threshold range can be associated with a feature or characteristic of an anatomical part to indicate the severity of the spinal deformity. In some embodiments, the threshold and / or threshold range can be determined based on a patient population. For example, the threshold and / or threshold range can be an average across image data obtained for various patient populations, such as an average across different ages, genders, etc. In some embodiments, the spine deformity can be classified (e.g., no deformity, mild deformity, moderate deformity, severe deformity, etc.) based on evaluating whether one or more features or characteristics fall within a particular threshold range and / or value. For example, the spine analysis model may include a first range of spinal surface area values ​​indicative of no deformation, a second range of spinal surface area values ​​indicative of mild deformation, and a third range of spinal surface area values ​​indicative of severe deformation. If the analysis of the spinal surface area performed at 544 returns a surface area within the first range, the spine analysis model may output "no deformation". However, if the analysis of the spinal surface area performed at 544 returns a surface area within the second or third range, the spine analysis model may output "mild deformation" or "severe deformation", respectively. As another example, the spine analysis model may include a first threshold value of distance between adjacent vertebrae (e.g., an anterior distance, a posterior distance, a maximum distance, or a minimum distance) indicative of no deformation, a second threshold value of distance between adjacent vertebrae indicative of mild deformation, and a third threshold value of distance between adjacent vertebrae indicative of severe deformation. If the analysis of the vertebrae performed at 544 returns a distance lower than the first threshold, the spine analysis model may output "no deformation".However, if the vertebral analysis performed at 544 returns a distance greater than the second threshold or a distance greater than the third threshold, the spine analysis model can output "mild deformity" or "severe deformity", respectively.

[0141] In some embodiments, the spine analysis model can predict a spinal deformity assessment for the ROI by comparing the current anatomical feature analysis to the historical spine analysis data retrieved at 546. For example, if a comparison of the current anatomical feature values ​​to the historical data retrieved at 546 indicates a change in disc height (e.g., a decrease in disc height) within the ROI that is greater than a predetermined threshold or percentage, the spine analysis model can predict a possible spinal deformity.

[0142] Optionally, at 548, if all ROIs of the image data have not been processed (548:NO), process 540 may return to 541 and repeat 541-547. Once all ROIs have been processed, process 540 ends and may proceed to 580, as described in connection with FIG.

[0143] 7A and 7B, described below, provide examples of spinal stenosis assessment 550 and disc degeneration assessment 560.

[0144] a) Stenosis assessment Spinal stenosis is characterized by the narrowing of the spinal canal and / or intervertebral foramen.Spinal stenosis can be caused by soft tissue changes such as intervertebral disc herniation (e.g., displacement of intervertebral disc tissue, such as displacement of nucleus pulposus in intervertebral disc), fibrous scar or tumor, or can be caused by changes in bone structure (e.g., vertebral body, pedicle, transverse process, lamina, and / or spinous process), such as intervertebral disc collapse, osteophyte formation, spondylolisthesis (e.g., anterior slippage of vertebra).Thus, spinal stenosis can result in a decrease in the volume and / or change in shape of the spinal canal.

[0145] FIG. 14A provides an illustration of the anatomy of the spine under normal conditions 1302 and spinal stenosis 1304. As shown, in spinal stenosis, the volume and shape of the dural sac or sheath surrounding the spinal cord (and therefore the space for the spinal cord) is reduced. FIG. 14B shows a grading system for classification of spinal stenosis developed by Lee et al. ("Lee grading system"). The Lee grading system defines various qualitative metrics for classifying different cases of spinal stenosis. The Lee grading system is based on the obstruction of the cerebrospinal fluid (CSF) space and the distribution of nerve roots within the dural sac as viewed on MRI images (e.g., T2-weighted images). Under the Lee grading system, if the anterior CSF space is not obstructed, the case is classified as grade 0 (no stenosis, 1312). If the anterior CSF space is mildly obstructed but all nerve roots can be clearly separated from each other, the case is classified as grade 1 (mild stenosis, 1314). If the anterior CSF space is moderately obstructed and some of the nerve roots are clumped together (e.g., cannot be visually separated), the case is classified as grade 2 (moderate stenosis, 1316). If the anterior CSF space is severely obstructed, showing significant compression of the dural sac, and the nerve roots appear as a single bundle (e.g., none can be visually separated from one another), the case is classified as grade 3 (severe stenosis, 1318).

[0146] As mentioned above, the identification of spinal stenosis by surgeons is subjective and can be inconsistent. Even with grading systems such as the Lee grading system, different surgeons looking at the same MRI scan and using the same grading system can reach different conclusions and therefore recommend different clinical interventions. Thus, the systems, devices, and methods described herein can provide a more reliable assessment of spinal stenosis by taking a more quantitative approach (e.g., in assessing the dural sac and / or other anatomical structures around the spinal canal).

[0147] FIG. 7A is a flow chart of a stenosis assessment process 550, according to some embodiments. At 551, an ROI can be selected in the image data (e.g., similar to 531 in FIG. 6A and 541 in FIG. 6B). The ROI can include one or more anatomical parts of interest, such as one or more vertebrae and surrounding soft tissue (e.g., nerve roots) and / or other structures. At 552, a 2D image associated with the ROI can be selected (e.g., similar to 532 in FIG. 6A and 542 in FIG. 6B). These images can include an axial view of the patient's anatomy, a sagittal view of the patient's anatomy, a coronal view of the patient's anatomy, and / or combinations thereof. Optionally, at 553, these 2D images can be combined to generate a 3D volume or sub-volume for further analysis (e.g., similar to 533 in FIG. 6A and 543 in FIG. 6B).

[0148] At 554, one or more anatomical features of the patient's anatomy may be analyzed in the selected image data (e.g., similar to 544 of FIG. 6B). In some embodiments, at 555a, the spinal cord and / or thecal sac may be identified in the selected image data. The selected image data may include axial scans of a portion of the spine (e.g., axial scans of C1-C2, ...C6-C7, C7-TH1, TH1-TH2, TH2-TH3, ...TH11-TH12, TH12-L1, L2-L3, L4-L5, L5-S1, ...S6-S7, etc.). Identification of the spinal cord and / or thecal sac may be based on the output of the segmentation model at 520 or manual identification / labeling of the spinal cord and / or thecal sac. FIGS. 9A-9E show exemplary axial images of a portion of the patient's spine L5-S1. The axial images include segmentation data identifying the spinal cord 912a. Similarly, FIGS. 10A-10B show exemplary axial images of portions of the patient's spine L5-S1 and TH12-L1, respectively. The axial images include segmentation data with portions 1002 and 1022 of FIGS. 10A-10B representing the spinal cord. At 555b, the area of ​​the spinal cord and / or the thecal sac can be determined based on the image data and / or segmentation data. For example, the portions of the images representing the spinal cord and / or the thecal sac can be analyzed and the areas of these portions can be measured. In some embodiments, the surface area can be represented as a number of pixels, while in other embodiments, the surface area can be represented as an actual metric (e.g., mm 2) In some embodiments, when multiple axial scans of a portion of the spine are used to analyze the spinal cord and / or thecal sac, the cross-sectional surface area of ​​the spinal cord and / or thecal sac can be measured in each axial scan, and an average metric (e.g., mean, median, etc.) can be used to approximate the spinal cord and / or thecal sac surface area for that portion of the spine. The area of ​​the spinal cord and / or thecal sac can indicate the presence and / or severity of stenosis. For example, a smaller surface area associated with the spinal cord and / or thecal sac can indicate a greater severity of stenosis. Although surface area is described as being used to assess stenosis, it can be appreciated that other parameters associated with the spinal cord and / or thecal sac region can be used to assess stenosis, including, for example, the lateral dimensions (e.g., width, height, etc.) of the thecal sac region, the shape of the thecal sac region, etc.

[0149] Additionally or alternatively, at 556a, a nerve root can be identified in the selected image data. Identification of the nerve root can be based on the output of the segmentation model at 520 or manual identification / labeling of the nerve in the image data. For example, one of the segmentation data can be used to identify a lateral nerve root and an intracameral nerve root. The image data can include one or more axial, sagittal, and / or coronal images that provide a view of the nerve through an opening between adjacent vertebrae. At 556b, the nerve root can be tracked from a first portion of the nerve root to a second portion of the nerve root (e.g., from the lateral recess to the vertebral foramen). Tracking of the nerve root can be an indicator of the presence and / or severity of stenosis. For example, a discontinuity of the nerve root, a narrowed portion of the nerve root, etc. can indicate spinal canal stenosis. In some aspects, the location or severity of the stenosis can be evaluated based on one or more regions that include, for example, a discontinuity or narrowed portion of the nerve root.

[0150] At 557, the spine analysis model can be used to assess the presence and / or severity of stenosis. For example, the spine analysis model can output a prediction of whether there is stenosis and / or the severity of the stenosis based on the anatomical feature analysis performed at 554. For example, threshold ranges and / or thresholds can be associated with different classes of stenosis (e.g., no stenosis, mild stenosis, severe stenosis) and / or grades of stenosis (grades I-IV). As an example, if the surface area of ​​the spinal cord and / or thecal sac (e.g., as determined at 555b) is below a first threshold, the spine analysis model can classify the selected image and / or ROI as severe stenosis. However, if the surface area of ​​the spinal cord and / or thecal sac (e.g., as determined at 555b) is above the first threshold but below a second threshold, the spine analysis model can classify the selected image and / or ROI as mild stenosis. Alternatively, if the surface area of ​​the spinal cord and / or thecal sac (e.g., as determined at 555b) is above a second threshold, the spine analysis model can classify the selected image and / or ROI as having no stenosis. As another example, if the tracked nerve root (e.g., as tracked at 556b) indicates a pinched nerve root (e.g., with one or more discontinuities in the nerve root tracking and / or narrowing of the nerve root to a diameter that falls within a certain predetermined range), the spine analysis model can classify the selected image and / or ROI as having spinal stenosis.

[0151] At 558, the stenosis assessment (e.g., output of the spine analysis model at 557) can be associated with one or more vertebral levels, for example, based on the output of the level identification process 530 of FIG. 6A and / or manual identification of vertebral levels. As described above with reference to 530, the level identification model (e.g., trained CNN 300 of FIG. 3B) can assign vertebral level or order identifiers (e.g., C1-S5, or C1-C7, TH1-TH12, L1-L5 (L6), and / or S1-S5) to portions of the image data. The spine analysis model (e.g., output of the spine assessment process 540 of FIG. 6B and / or output of the spine analysis model at 557) can assign deformity assessments (e.g., no stenosis, mild stenosis, and / or severe stenosis) to selected images and / or ROIs based on anatomical feature analysis at 554. The spine deformity assessments can be associated with associated vertebral levels at 558. For example, in Figure 9A, the sequence identifier "L5-S1" is associated with the deformation assessment "mild stenosis" (e.g., assessed based on the surface area of ​​the spinal cord). The stenosis assessment can then end.

[0152] Although not shown as an iterative process, the stenosis assessment process 550 of FIG. 7A may be repeated from 551-558 if additional ROIs and / or anatomical portions need to be analyzed for stenosis.

[0153] In some embodiments, the quantitative assessment of spinal stenosis can be performed using morphometric quantitative measurements including, for example, the dural sac compression factor. FIG. 15 is a flow chart of a stenosis assessment process 1450, according to some embodiments. The stenosis assessment process 1450 can be used in combination with or instead of the stenosis assessment process 550 described above in connection with FIG. 7A. As shown in FIG. 15, the stenosis assessment process can include selecting an axial volume of the patient's dural sac (or disc block including the dural sac) for analysis at 1451. The axial volume can be, for example, an axial volume obtained via the level identification process 1130 described in connection with FIG. 12 above. At 1454, one or more anatomical features of the dural sac and / or surrounding anatomical structures can be analyzed. In some embodiments, such analysis can include determining a compression factor of the dural sac. To determine the compression factor, the dural sac surface area of ​​each 2D image within the axial volume can be determined at 1454a, and the following calculations are performed:

number

[0154] In particular, the process of determining the dural sac compression factor may include calculating the average dural sac surface area of ​​the first and last images in the axial volume at 1454b, identifying the smallest or smallest dural sac surface area (or a set of smaller dural sac surface areas) among the remaining images (i.e., the middle images of the axial volume) at 1454c, and then comparing the smallest dural sac surface area to the average dural sac surface area of ​​the first and last images at 1454d (e.g., by calculating a ratio of 2). Typically, dural sac compression occurs in the middle of the intervertebral level, and with age and wear, the intervertebral disc may bulge and cause pressure on the dural sac. Conversely, the border of the dural sac captured by the border images (i.e., the first and last images of the axial volume) is usually undamaged or damaged, since it is taken on or near the adjacent bony structures. Therefore, the average dural sac surface area for the first and last images of the axial volume can provide a good approximation of the expected surface area for the more central portion of the dural sac (as represented by the middle images of the axial scan). Therefore, equation 1 as above can be used to provide a quantitative indication of how much the dural sac surface area at the smallest intervertebral level differs from the expected dural sac surface area.

[0155] At 1456, the compression coefficient of the dural sac can be compared to the compression coefficient of a patient population. In some embodiments, the patient population can be a general patient population. Alternatively, the patient population for comparison can be one that shares similar attributes as the patient being tested, for example, by belonging to the same age group and / or by having the same gender, height, weight, ethnicity, and / or other shared attributes. Comparison with a patient population can provide objective data to the clinician without providing a direct assessment of the classification of the stenosis status.

[0156] Optionally, at 1457, a grade or classification can be assigned to the patient's spinal condition, for example, based on the Lee grading system. For example, different thresholds can be set that correlate different ranges of compression coefficients to grades or classifications of the Lee grading system. In some embodiments, the different thresholds can be set based on an analysis of compression coefficients of the dural sacs associated with known Lee grades. For example, a training data set can be obtained that includes axial volumes of a plurality of dural sacs with different degrees or severity of stenosis, and each of these dural sacs can be associated with a known Lee grade. A compression coefficient for each of the axial volumes can be determined (e.g., by a computing device described herein), and then a correlation (e.g., linear or nonlinear regression or other relationship) can be derived between the compression coefficient and the Lee grade based on the Lee grade for each of the axial volumes. Such correlation can then be used to set different thresholds or ranges corresponding to each Lee grade.

[0157] At 1458, the stenosis analysis, including the compression index analysis and / or the classification of stenosis, can be associated with one or more levels of the spine. For example, the stenosis analysis can be associated with an ordinal identifier of the two vertebrae that join the disc (e.g., L5-S1, L4-L5, L3-L4, L2-L3, L1-L2, T12-L1, etc.). Although not shown, such an analysis can also be stored, for example, in a memory.

[0158] If all axial volumes have been processed (1459: YES), the process may end. If additional axial volumes still need to be processed (1459: NO), the process may repeat the steps with another axial volume at a different level. Once all of the axial volumes of the dural sac have been analyzed, the process may continue to visualize the stenosis analysis. For example, in some embodiments, the process may proceed to 580, as previously described with reference to the high level flow shown in FIG. 5.

[0159] b) Evaluation of intervertebral disc degeneration Disc degeneration can be caused by dehydration of the annulus fibrosus and / or dehydration of the nucleus, which can result in loss of the structural and functional integrity of the disc.

[0160] FIG. 19 provides an illustration of the spinal anatomy under normal conditions (left) and degenerative disc disease (right). As shown, in degenerative disc disease, the nucleus is dehydrated, which can lead to endplate calcification and reduced nutrient transport through the tissue. FIG. 20A shows a grading system for classification of disc degeneration developed by Pfirrmann et al. ("Pfirrmann grading system"). The Pfirrmann grading system defines various qualitative metrics for classifying different cases of disc degeneration. The classification includes five grades based on the homogeneity of the disc space and the brightness of the signal intensity in MRI images (e.g., T2-weighted images). In the Pfirrmann grading system, the following qualitative characteristics are associated with each grade: · Grade I (Example A in Fig. 20A): The disc structure is homogeneous, with bright hyperintense white signal intensity and normal disc height. Grade II (Example B in Fig. 20A): The disc structure is heterogeneous with a very intense white signal. The distinction between the nucleus and the annulus is clear, the disc height is normal, and there may or may not be a horizontal gray band. Grade III (Example C in Fig. 20A): The disc structure is heterogeneous with intermediate grey signal intensity. The distinction between the nucleus and the annulus fibrosus is unclear, and the disc height is normal or slightly reduced. Grade IV (Example D in Fig. 20A): The disc structure is heterogeneous with low to dark grey signal intensity. The distinction between the nucleus and annulus fibrosus is lost, and the disc height is normal or moderately reduced. Grade V (Example E in Fig. 20A): The structure of the disc is heterogeneous, with low-intensity black signal intensity. The distinction between the nucleus and the annulus is lost and the disc space collapses.

[0161] FIG. 20B illustrates an exemplary flow for assigning a Pfirrmann grade to a disc, for example, based on viewing an MRI scan of the disc. At 1802, the physician can consider whether there is a uniform bright white structure. If so, the disc can be classified as grade I. If no, at 1804, the physician can consider whether there is a non-uniform white structure, with or without one or more horizontal bands. If so, the disc can be classified as grade II. If not, at 1806, the physician can consider whether there is a clear distinction between the annulus fibrosus and the nucleus. If so, the disc can be classified as grade III. If not, the physician can consider whether the disc space has collapsed. If no, the disc can be classified as grade IV. If so, the disc can be classified as grade V.

[0162] Nevertheless, the identification of disc degeneration by surgeons can be subjective and inconsistent. Even with grading systems such as the Pfirrmann grading system, different surgeons looking at the same MRI scan and using the same grading system can reach different conclusions and therefore recommend different clinical interventions. Thus, the systems, devices, and methods described herein can provide a more reliable assessment of disc degeneration by taking a more quantitative approach (e.g., in assessing the rate of degeneration and / or the distance between adjacent discs).

[0163] FIG. 7B is a flow chart of a disc degeneration assessment process 560, according to some embodiments. At 561, an ROI can be selected in the image data (e.g., similar to 531 in FIG. 6A and 541 in FIG. 6B). The ROI can include one or more anatomical portions of interest, such as one or more vertebrae and surrounding discs and / or other structures. At 562, 2D images associated with the ROI can be selected (e.g., similar to 532 in FIG. 6A and 542 in FIG. 6B). These images can include axial views of the patient's anatomy, sagittal views of the patient's anatomy, coronal views of the patient's anatomy, and / or combinations thereof. Optionally, at 563, these 2D images can be combined to generate a 3D volume or sub-volume for further analysis (e.g., similar to 533 in FIG. 6A and 543 in FIG. 6B).

[0164] At 564, one or more anatomical features of the patient's anatomy may be analyzed in the selected image data (e.g., similar to 544 in FIG. 6B). In some embodiments, at 565a, intervertebral discs may be identified in the selected image data, e.g., based on the output of the segmentation model at 520 or manual identification / labeling of intervertebral discs. As an example, the selected image data may be axial images of a portion of a spine (e.g., axial scans of C1-C2, ...C6-C7, C7-TH1, TH1-TH2, TH2-TH3, ....TH11-TH12, TH12-L1, L2-L3, L4-L5, L5-S1, ...S6-S7, etc.). A portion of the intervertebral disc (e.g., annulus fibrosus and / or nucleus) may be segmented in the axial images by the segmentation model. FIGS. 9A-9E show exemplary axial scans and / or axial images of a portion of a patient's spine L5-S1 with the intervertebral disc segmented in the images. As shown in FIG. 9C, 914 represents the disc annulus and 914' represents the nucleus. At 565b, the shape, area, or other geometric characteristics of the disc and / or its components (e.g., disc, annulus, and / or nucleus) can be determined from the selected images. The shape and / or area of ​​the disc can indicate the presence and / or severity of disc degeneration. For example, a narrower or smaller area of ​​the nucleus (e.g., 914' in FIG. 9C) can indicate more severe disc degeneration.

[0165] Additionally or alternatively, at 566a, upper and lower vertebrae surrounding the intervertebral disc may be identified in the selected image data, for example, based on the output of the segmentation model at 520 or manual identification / labeling of the vertebrae. At 566b, the distance between the upper (upper) and lower (lower) vertebrae of the intervertebral disc (i.e., two adjacent vertebrae of the intervertebral disc) may be determined. This distance may indicate the presence and / or severity of disc degeneration. In particular, disc degeneration may result in a reduction in the height of the intervertebral disc, thereby reducing the distance between the upper and lower vertebrae. Thus, the distance between the upper and lower vertebrae may indicate the presence and / or severity of disc degeneration.

[0166] At 567, the spine analysis model can be used to assess the presence and / or severity of disc degeneration. For example, the spine analysis model can output a prediction of whether there is disc degeneration and / or the severity of disc degeneration based on the anatomical feature analysis performed at 564. For example, a threshold range and / or threshold can be associated with the severity of disc degeneration. As an example, if the height of the disc and / or the distance between two adjacent vertebrae is below a threshold, the spine analysis model can classify the selected image and / or ROI as having disc degeneration. Similarly, if the area of ​​the disc nucleus (e.g., 914' in FIG. 9C) is below a threshold, the spine analysis model can classify the selected image and / or ROI as having disc degeneration.

[0167] At 568, the disc degeneration assessment (e.g., the output of the spine analysis model at 567) can be associated with one or more vertebral levels, for example, based on the output of the level identification process 530 of FIG. 6A and / or manual identification of vertebral levels. As described above with reference to 530, the level identification model (e.g., the trained CNN 300 of FIG. 3B) can assign vertebral level or order identifiers (e.g., C1-S5, or C1-C7, TH1-TH12, L1-L5 (L6), and / or S1-S5) to portions of the image data. The spine analysis model (e.g., the output of the spine assessment process 540 of FIG. 6B and / or the output of the spine analysis model at 567) can assign deformity assessments (e.g., no disc degeneration, mild disc degeneration, and / or severe disc degeneration) to selected images and / or ROIs based on anatomical feature analysis at 564. The spine deformity assessments can be associated with associated vertebral levels at 568. The disc degeneration assessment can then be completed.

[0168] Although not shown as an iterative process, the disc degeneration assessment process 560 of FIG. 7B can be repeated from 561-568 if additional ROIs and / or anatomical portions need to be analyzed for disc degeneration.

[0169] In some embodiments, quantitative assessment of disc degeneration can be performed using morphometric quantitative measurements including, for example, disc degeneration ratios. FIG. 21 is a flow chart of a disc degeneration assessment process 2060 according to an embodiment. The disc degeneration assessment process 2060 can be used in combination with or instead of the disc degeneration process 560 described above in connection with FIG. 7B. As shown in FIG. 21, the process 2060 can include selecting a sagittal image (or a set of sagittal images or sagittal volumes) of the patient for analysis at 2061. The sagittal images can show a range of intervertebral levels suitable for evaluating different regions of the spine, for example. For example, to evaluate the lower back, a sagittal image showing the range S1-T12 can be selected. At 2062, a first disc shown in the sagittal image can be selected for analysis. In some embodiments, segmentation data and / or level identification data of the sagittal images can be loaded and such data can be used to identify each intervertebral level. The intervertebral levels can then be assessed individually, for example, by starting at one end of the intervertebral range (eg, S1) and progressing to the other end of the intervertebral range (eg, T12).

[0170] At 2064, one or more anatomical features of the disc and / or surrounding anatomical structures can be analyzed. In some embodiments, such analysis can include determining a disc degeneration ratio. To determine the disc degeneration ratio, the mean intensity and actual volume of the annulus fibrosus of the disc can be determined at 2064a, the mean intensity and actual volume of the nucleus of the disc can be determined at 2064b, and then the following calculations can be performed at 2064c:

number

[0171] As represented by Equation 2, the disc degeneration ratio or factor is based on calculations of the volume and intensity of the annulus and nucleus. These calculations can be based on sagittal image data and segmentation data. For each intervertebral level, the volume and average intensity of the nucleus and annulus can be determined and used in Equation 2 to determine the disc degeneration ratio.

[0172] Similar to the compression index analysis described above in connection with FIG. 15, the disc degeneration ratio can be compared to the disc degeneration ratio of a patient population at 2066. In some embodiments, the patient population can be a general patient population. Alternatively, the patient population for comparison can be one that shares similar attributes as the patient being examined, for example, by belonging to the same age group and / or by having the same gender, height, weight, ethnicity, and / or other shared attributes. Comparison with a patient population can provide objective data to the clinician without providing a direct assessment of the classification of the disc degeneration status.

[0173] Optionally, at 2067, a grade or classification can be assigned to the patient's spinal condition, for example, based on the Pfirrmann grading system. For example, different thresholds can be set that correlate different ranges of disc degeneration ratios with Pfirrmann grading system grades or classifications. In some embodiments, the process for assigning such thresholds can be similar to the classification process described with respect to the spinal stenosis analysis above, at 1457 of process 1450 shown in FIG. 15.

[0174] At 2068, a disc degeneration analysis, including a disc degeneration ratio analysis and a classification of disc degeneration, can be associated with the disc. For example, the disc degeneration analysis can be associated with an ordinal identifier of the two vertebrae that join the disc (e.g., L5-S1, L4-L5, L3-L4, L2-L3, L1-L2, T12-L1, etc.). Although not shown, such an analysis can also be stored, for example, in a memory.

[0175] If all discs have been analyzed (2069: YES), the process may end. If more discs still need to be analyzed (2069: NO), the process may repeat the steps with another disc. Once all of the discs in one or more sagittal images have been analyzed, the process may continue to visualize the disc degeneration analysis. For example, in some embodiments, the process may proceed to 580, as described above with reference to the high-level flow shown in FIG. 5.

[0176] 5. Visualization of Spine Analysis 8 illustrates an example report 800 generated by the systems and methods described herein, according to some embodiments. The report 800 can include patient information 802, such as age, past spinal deformity information, etc. A spinal deformity diagnosis 804 can be generated based on a level identification process (e.g., level identification process 530 of FIG. 6A) and a spinal deformity assessment process (e.g., spinal deformity assessment process 540 of FIG. 6B, stenosis assessment process 550 of FIG. 7A, and / or disc degeneration assessment process 560 of FIG. 7B). For example, the level identification process can generate vertebral level or order identifiers (e.g., C1-S5, or C1-C7, TH1-TH12, L1-L5 (L6), and / or S1-S5), such as a first level 806a and a second level 806b in the report 800. The spinal deformity assessment process can generate a spinal deformity diagnosis, such as diagnosis 808a and diagnosis 808b in the report (e.g., no stenosis, mild stenosis, moderate stenosis, severe stenosis, no disc degeneration, mild disc degeneration, moderate disc degeneration, severe disc degeneration, no deformation, mild deformation, and / or severe deformation). The output of the level identification process can be associated with the output of the spinal deformity process. For example, a first level 806a can be associated with a diagnosis 808a, thereby indicating that a diagnosis was performed on the first level using image data. Similarly, a second level 806b can be associated with a diagnosis 808b.

[0177] In some embodiments, the report 800 can include output of an anatomical feature analysis 810 (e.g., analysis 544 of FIG. 6B, analysis 554 of FIG. 7A, and analysis 564 of FIG. 7B). For example, the report 800 can include surface area of ​​the spinal cord and / or thecal sac, area of ​​the intervertebral disc, distance between adjacent vertebrae, etc. The anatomical feature analyses, such as 812a and 812b of FIG. 8, can also be associated with vertebral level or order identifiers, such as 806a and 806b.

[0178] In some embodiments, the report 800 may include a virtual representation of the patient's anatomy 820. For example, the report 800 may include a 3D and / or 2D view of the patient's anatomy. In some embodiments, these 3D and / or 2D views of the patient's anatomy may be used to provide virtual or augmented reality images for display by a computer-assisted surgery system. In such a system, a virtual 2D or 3D view of the patient's anatomy may be displayed over an actual portion of the patient's anatomy (e.g., a surgical site). In some embodiments, such a virtual or augmented reality representation of the patient's anatomy may include information 822 regarding a spinal deformity diagnosis (e.g., diagnosis 808a and 808b) and / or anatomic feature analysis (e.g., analysis 812a and 812b).

[0179] 9A-9E show a portion of an example report 900 generated for a patient with mild spinal stenosis. Although this portion of the example report is shown in a separate figure, it can be understood that all portions (or one or more portions) of the report can be presented together, for example, on a single page, screen, or other interface. The report includes a spinal deformity diagnosis 904 (e.g., similar to spinal deformity diagnosis 804). A vertebral level or order identifier (e.g., similar to levels 806a and 806b) "L5-S1" is associated with a spinal deformity diagnosis (e.g., similar to diagnosis 808a and 808b) "Mild Spinal Stenosis." The report can include several images of anatomical structures and / or ROIs. An anatomical feature analysis (e.g., similar to analysis 812a and 812b) can be performed on each image. The analysis of each image can be listed as 911. Additionally, the analysis of each image (e.g., similar to analyses 812a and 812b) can be associated with a vertebral level or order identifier (e.g., similar to levels 806a and 806b). For example, for image number 0, the analysis "Spinal Cord Surface Area: 126.78 mm2" may be associated with a vertebral level or order identifier (e.g., similar to levels 806a and 806b). 2" 912a may be associated with the ordinal identifier "L5-S1." In FIGS. 9A-9E, the anatomical feature analysis performed is of the spinal cord, however, it should be readily understood that any suitable analysis may be performed to determine spinal deformity. For example, compression index, disc 914 area and / or nucleus 914' area may be determined to diagnose disc degeneration in a patient.

[0180] 10A illustrates a diagnosis of severe stenosis in a patient. Left image 1000 shows a scan 1010 of the patient's anatomy, and right image 1000' shows scan 1010 with segmentation data overlaying the scan. An anatomical feature analysis of scan 1010 (e.g., similar to analyses 812a and 812b) reveals a 56.25 mm stenosis. 2 As previously discussed, a very narrow spinal cord region may indicate severe stenosis. This analysis is associated with an ordinal identifier (e.g., similar to levels 806a and 806b) "L5-S1." FIG. 10B illustrates an example of a patient with a diagnosis of no stenosis. Left image 1020 shows scan 1010' of the patient's anatomy, and right image 1020' shows scan 1010' with segmentation data overlaying the scan. An anatomical feature analysis of scan 1010' (e.g., similar to analyses 812a and 812b) yields a spinal cord area of ​​288.41 mm. 2 This analysis is associated with an ordinal identifier (e.g., similar to levels 806a and 806b) "TH12-L1."

[0181] In some embodiments, the systems and methods described herein can generate a stenosis report that includes information related to an analysis of one or more axial volumes associated with one or more levels of the spine. FIG. 16 illustrates an example stenosis report 1500 according to an embodiment. The stenosis report 1500 can include elements similar to those of the report 800. For example, the stenosis report 1500 can include patient information 1502. The stenosis report 1500 can also include a scanned image 1504, such as a 2D image or scan, of the patient's anatomy. In some embodiments, the 2D image can optionally be depicted with segmentation data 1505 and / or level identification data 1506. For example, the 2D image can include segmentation data, such as labels identifying different anatomical parts of interest, and / or level identification data 1506, such as labels identifying different vertebral or intervertebral levels. In some embodiments, the 2D image can be a particular image selected from the 3D volumetric image data that shows a more severe stenosis case (e.g., Lee grade 2 of grade 3). The 2D images can be automatically selected by a computing device (e.g., computing device 110, 210) when analyzing one or more image sets, or the 2D images can be manually selected by the physician after viewing the stenosis analysis. In some embodiments, multiple 2D images can be depicted in the stenosis report 1500, while in other embodiments, a single 2D image is depicted at a time. In some embodiments, a spatial sequence of 2D images can be shown, such as, for example, all images within an axial volume that have been determined to have a stenosis or classified as having a more severe case of stenosis.

[0182] The stenosis report 1500 may also include a stenosis analysis 1510. The stenosis analysis may include dural sac surface area data 1511 and / or compression index data 1512. The dural sac surface area data 1511 may include, for example, a minimum dural sac surface area 1511a (e.g., a minimum dural sac surface area or a smaller set of dural sac surface areas of one or more axial volumes that include the dural sac), a dural sac surface area over a set of scans 1511b (e.g., a dural sac surface area for all or a subset of scans in one or more axial volumes that include the dural sac, an average ... The dural sac surface area may include a range of dural sac surface area for all or a subset of scans, and / or other quantitative indicators of dural sac surface area across multiple scans), a comparison of the patient's dural sac surface area to an average of a population 1511c (e.g., a percentile of the patient's dural sac surface area for one or more levels of the spine compared to the dural sac surface area of ​​the population), and / or a grade or classification of the stenosis based on the dural sac surface area 1511d (e.g., a Lee grade for one or more dural sacs based on the dural sac surface area of ​​their respective axial volumes).

[0183] The compression coefficient data 1512 may include a comparison of the patient's compression coefficient to values ​​of the population 1513a (e.g., percentiles of the patient's compression coefficient for one or more thecal sacs compared to the compression coefficients of the population), and / or a grade or classification of the stenosis based on the compression coefficient 1513b (e.g., Lee grade for one or more thecal sacs based on the compression coefficients of their respective axial volumes).

[0184] The quantitative metrics 1511a, 1511b, 1511c, 1511d, 1513a, 1513b and / or other metrics included in the report can be presented in textual, graphical (e.g., plots, bar graphs, etc.) and / or other suitable formats. In some embodiments, the stenosis report 1500 can be interactive, and a user (e.g., a physician) can select from a menu of options to update the report and / or reorganize the content within the report. For example, a user can choose to hide one or more metrics while displaying other metrics. Or, a user can choose to display a first set of 2D scans and then a second set of 2D scans. In some embodiments, a user can choose to reduce or increase the size of certain information presented. For example, a user can choose to focus on a portion of a 2D scan and / or to zoom in or out on a graph or plot. A user can select what is presented using a keyboard, mouse, touch screen, audio device, and / or other input device.

[0185] 17A-17D are diagrams illustrating a portion of an exemplary stenosis report, according to an embodiment. Although this portion of the stenosis report is shown in a separate figure, it can be understood that all portions (or one or more portions) of the stenosis report can be presented together, for example, on a single page, screen, or other interface. The stenosis report can include patient information 1602, such as the patient's name, sex, age, and date of birth, as well as information about the patient's physician and the patient's scan (e.g., scan date, scan time, scanner model, etc.). The stenosis report can also include scan images 1604a, 1604b. Scan image 1604a can be an axial scan showing the smallest dural sac area corresponding to the area of ​​high compression. Scan image 1604b can be a sagittal scan showing the entire region of interest of the spine (e.g., L5-S1). Segmentation data 1605a, 1605b can also be shown in each scan image 1604a, 1604b. The stenosis report may also include a table 1610 showing quantitative metrics related to the stenosis analysis. The table 1610 may identify the intervertebral levels being analyzed (e.g., L5-S1, L4-L5, etc.) and the corresponding minimum dural sac surface area and compression coefficient for each level.

[0186] The stenosis report may also include a comparison graph 1611 showing the thecal sac surface area for each intervertebral level across a scan of the axial volume for each intervertebral level. Each of the graphs 1611 may correspond to a different intervertebral level. In some embodiments, the graphs 1611 may include color coding to indicate when a patient's thecal sac surface area falls into different classifications or grades of stenosis based on a population metric. For example, FIGS. 18A and 18B show two detailed views of a comparison graph, according to an embodiment. FIG. 18A shows the thecal sac surface area (mm) across an axial scan for the L5-S1 intervertebral levels. 2 FIG. 18B shows a comparison graph 1710 illustrating the dural sac surface area (in mm 2A comparison graph 1720 showing the dural sac surface area (in units) of the stenosis is shown. For each intervertebral level, different ranges or thresholds can be defined that are associated with different classifications of stenosis. As previously described under the stenosis evaluation process, these ranges or thresholds can be set based on the dural sac surface area of ​​a population with known stenosis grades or classifications. Table 1702 shows examples of thresholds that can be set for each of the stenosis grades or classifications. The thresholds for each intervertebral level are different because stenosis may present itself in a different way at each level, as shown in table 1702, and therefore a population analysis must be performed at each level to obtain clinical thresholds for classifying stenosis at each level. The graphs 1710, 1720 can then show the ranges for each grade or classification of stenosis in different colors, i.e., green for no stenosis, yellow for mild stenosis, orange for moderate stenosis, and red for severe stenosis.

[0187] 17A-17D, the stenosis report may also include a graphic 1612 for comparing compression coefficients across different intervertebral levels. As shown, the graphic 1612 may include a value of 1 as an upper limit, which corresponds to no compression. The graphic 1612 may include a compression coefficient associated with the 2.5th percentile of the pathological group for each level as a lower limit. For example, for the L5-S1 level, the lower limit may be 0.23, which corresponds to the compression coefficient of the 2.5th percentile of the pathological group. Similarly, the level L4-L5 may include a value of 0.31 as its lower limit, which corresponds to the compression coefficient of the 2.5th percentile of the pathological group. Each of the lower limits represents significant compression, and thus may provide an objective comparison for a physician to assess the severity of a patient's spinal stenosis at each of the patient's intervertebral levels. In this example, the 2.5 percentile is used as the lower limit, but it can be understood that other lower limits can be used, such as about the 1st percentile, about the 5th percentile, about the 10th percentile, about the 20th percentile, including all values ​​and ranges therebetween.

[0188] In some embodiments, the systems and methods described herein can generate a disc degeneration report that includes information related to an analysis of one or more discs in a sagittal image or volume. FIG. 22 illustrates an exemplary disc degeneration report 2100, according to an embodiment. The disc degeneration report 2100 can include elements similar to those of the reports 800, 1500 described above. For example, the disc degeneration report 2100 can include patient information 2102. The disc degeneration report 2100 can also include a scanned image 2104, such as, for example, a 2D image or scan, of the patient's anatomy. In some embodiments, the 2D image can include one or more sagittal images, or portions thereof. For example, the systems and devices described herein can autonomously select and present a sagittal image that provides the best or better view of the disc of interest. In some embodiments, a close-up of a portion of the sagittal image showing one or more intervertebral levels with more severe disc degeneration can be shown. Additionally or alternatively, the display of the scan image can be interactive, and the physician can select to zoom in or focus on a portion of the sagittal image that shows one or more intervertebral levels. In some embodiments, the 2D image or scan can optionally be depicted with segmentation data 2105 and / or level identification data 2106. For example, the 2D image can include segmentation data, such as labels that identify different anatomical parts of interest, and / or level identification data 1506, such as labels that identify different vertebral or intervertebral levels.

[0189] The disc degeneration report 2100 may also include a disc degeneration analysis 2120. The disc degeneration analysis may include degeneration ratio data 2121 for one or more discs, as well as annulus volume 2122b, annulus mean intensity 2122b, nucleus volume 2123a, and nucleus mean intensity 2123b data used to determine the degeneration ratio for one or more discs. The disc degeneration report 2100 may also include a comparison of the patient's disc degeneration ratio and / or other metrics (e.g., distance between adjacent vertebrae) to the disc degeneration ratio of a population 2124. In some embodiments, such a comparison may be a visual showing the percentile of the patient's disc degeneration ratio within the larger population data. Alternatively or additionally, a numerical percentile of the patient's disc degeneration ratio relative to the population data may be provided.

[0190] The quantitative metrics 2121, 2122a, 2122b, 2123a, 2123b, and / or other metrics included in the report may be presented in textual, graphical (e.g., plots, bar graphs, etc.), and / or other suitable formats. Like report 1500, disc degeneration report 2100 may be interactive, for example, by allowing a user to adapt or update information presented in the report.

[0191] 23 illustrates an example of a disc degeneration report, according to an embodiment. The disc degeneration report may include patient information 2202, such as, for example, the patient's name, sex, age, and date of birth, as well as information about the patient's physician and the patient's scan (e.g., scan date, scan time, scanner model, etc.). The disc degeneration report may also include scan image and segmentation data, for example, similar to that described in FIGS. 17A-17D above. The disc degeneration report may also include a table 2220 that includes specific disc degeneration analysis data, including, for example, annulus volume at CC, nucleus volume at CC, and disc degeneration ratio 2221.

[0192] Although separate spinal stenosis and disc degeneration reports are described, it can be understood that different elements of each can be combined into a single report. In some embodiments, the physician can also interactively update the report to show, for example, spinal stenosis analysis information, then disc degeneration analysis information, and / or to show the different pieces of information side-by-side with each other.

[0193] While various inventive embodiments have been described and illustrated herein, those of ordinary skill in the art will readily envision various other means and / or structures for performing the functions and / or obtaining the results and / or one or more advantages described herein, and each such variation and / or modification is deemed to be within the scope of the inventive embodiments described herein. More generally, those of ordinary skill in the art will readily appreciate that all parameters, dimensions, materials, and configurations described herein are meant to be exemplary, and that the actual parameters, dimensions, materials, and / or configurations will depend on the particular application or applications for which the teachings of the present invention are used. Those of ordinary skill in the art will recognize, or be able to ascertain using no more than routine experimentation, many equivalents to the specific inventive embodiments described herein. Accordingly, it should be understood that the foregoing embodiments are presented by way of example only, and are within the scope of the appended claims and their equivalents. The inventive embodiments may be practiced otherwise than as specifically described and claimed. The inventive embodiments of the present disclosure relate to each individual feature, system, article, material, kit, and / or method described herein. Furthermore, any combination of two or more such features, systems, articles, materials, kits, and / or methods is included within the inventive scope of the present disclosure, if such features, systems, articles, materials, kits, and / or methods are not mutually inconsistent.

[0194] Also, various inventive concepts may be embodied as one or more methods, examples of which are provided. The acts performed as part of a method may be ordered in any suitable manner. Thus, while shown as sequential acts in the exemplary embodiments, embodiments can be constructed in which acts are performed in an order different from that shown, which may include performing some acts simultaneously.

[0195] As used herein, the terms "about" and / or "approximately," when used in conjunction with numerical values ​​and / or ranges, generally refer to a numerical value and / or range that is close to the recited numerical value and / or range. In some cases, the terms "about" and "approximately" may mean within ±10% of the recited value. For example, in some cases, "about 100 [units]" may mean within ±10% of 100 (e.g., from 90 to 110). The terms "about" and "approximately" may be used interchangeably.

[0196] Any and all references to publications or other documents, including but not limited to patents, patent applications, articles, web pages, books, and the like, presented anywhere in this application are incorporated herein by reference in their entirety. Furthermore, all definitions defined and used herein should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and / or ordinary meanings of the defined terms.

[0197] Some embodiments and / or methods described herein can be implemented by different software modules (executed on hardware), hardware, or a combination thereof. Hardware modules can include, for example, general-purpose processors, field programmable gate arrays (FPGAs), and / or application specific integrated circuits (ASICs). Software modules (executed on hardware) can be implemented in a variety of languages, including C, C++, Java, TM , Python, Ruby, Visual Basic TMThe programming language may be expressed in a variety of software languages ​​(e.g., computer code), including, for example, IDE ...

Claims

1. 1. A processor-implemented method comprising: selecting a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae; Identifying the spinal cord or dural sac in the ROI; determining one or more parameters associated with the spinal cord or thecal sac in the ROI, wherein the image data of the ROI includes a plurality of axial scans of vertebrae from the plurality of vertebrae, each axial scan of the plurality of axial scans including a cross section of the spinal cord or thecal sac adjacent to the vertebra; determining a surface area of ​​the cross-section of the spinal cord or dural sac in each axial scan from the plurality of axial scans; determining an average surface area of ​​the spinal cord or dural sac based on (1) the surface area of ​​the cross section determined for a first axial scan from the plurality of axial scans corresponding to a most superior scan of the vertebra; and (2) the surface area of ​​the cross section determined for a second axial scan from the plurality of axial scans corresponding to a most inferior scan of the vertebra; identifying one or more surface areas from among remaining surface areas of the cross-section determined for the plurality of axial scans excluding the first axial scan and the second axial scan; the one or more surface areas are the smallest or smallest of the remaining surface areas; identifying the one or more surface areas; determining a compression factor based on the one or more surface areas and the average surface area, wherein at least one of the one or more parameters comprises the compression factor; determining the one or more parameters, assessing the severity of spinal stenosis in the ROI based on the one or more parameters associated with the spinal cord or dural sac; A method comprising:

2. 2. The method of claim 1, wherein identifying the spinal cord or the dural sac comprises processing the image data of the ROI using a convolutional neural network (CNN) trained to segment the plurality of vertebrae and tissue structures surrounding the plurality of vertebrae to obtain segmentation data that identifies the spinal cord or the dural sac.

3. The method of claim 1 , wherein the one or more parameters include a cross-sectional area of ​​the spinal cord or dural sac.

4. The method of claim 1 , further comprising generating a plot of the surface area determined for the cross-sections in the multiple axial scans.

5. The method of claim 4 , wherein the plot includes one or more indicators marking ranges of cross-sectional area of ​​the spinal cord or dural sac associated with a plurality of different grades of stenosis.

6. The method of claim 1 , wherein determining the compression factor comprises determining a ratio of (1) an average of the one or more surface areas to (2) the average surface area.

7. 2. The method of claim 1, wherein assessing the severity of spinal stenosis in the ROI comprises assessing a grade of spinal stenosis based on the compression index, the grade being associated with one of no stenosis, mild stenosis, moderate stenosis, or severe stenosis.

8. Memory and a processor operatively coupled to the memory, selecting a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae; Identifying the spinal cord or the dural sac in the ROI; determining one or more parameters associated with the spinal cord or the dural sac in the ROI, wherein the image data of the ROI includes a plurality of axial scans of vertebrae from the plurality of vertebrae, each axial scan of the plurality of axial scans including a cross section of the spinal cord or the dural sac adjacent to the vertebra; and, to determine the one or more parameters of the spinal cord or the dural sac, the processor: determining a surface area of ​​the cross section of the spinal cord or dural sac in each axial scan from the plurality of axial scans; determining an average surface area of ​​the spinal cord or dural sac based on (1) the surface area of ​​the cross section determined for a first axial scan from the plurality of axial scans corresponding to a most superior scan of the vertebra; and (2) the surface area of ​​the cross section determined for a second axial scan from the plurality of axial scans corresponding to a most inferior scan of the vertebra; identifying one or more surface areas from among remaining surface areas of the cross section determined for the plurality of axial scans excluding the first axial scan and the second axial scan; the one or more surface areas are the smallest or smallest of the remaining surface areas; identifying the one or more surface areas; determining a compression factor based on the one or more surface areas and the average surface area, wherein at least one of the one or more parameters comprises the compression factor; and further configured to: assessing the severity of spinal stenosis in the ROI based on the one or more parameters associated with the spinal cord or dural sac; the processor configured to An apparatus comprising:

9. 9. The apparatus of claim 8, wherein the processor is configured to identify the spinal cord or the dural sac by processing the image data of the ROI using a convolutional neural network (CNN) trained to segment the plurality of vertebrae and tissue structures surrounding the plurality of vertebrae to obtain segmentation data that identifies the spinal cord or the dural sac.

10. The device of claim 8 , wherein the one or more parameters include a cross-sectional area of ​​the spinal cord or dural sac.

11. The apparatus of claim 10 , wherein the processor is further configured to generate a plot of the surface area determined for the cross-sections in the multiple axial scans.

12. A non-transitory processor-readable medium storing code representing instructions to be executed by a processor, the code causing the processor to: selecting a region of interest (ROI) within image data of a three-dimensional (3D) volume of a plurality of vertebrae of a spine, the ROI including image data of one or more vertebrae from the plurality of vertebrae and tissue structures surrounding the one or more vertebrae; Identifying the spinal cord or the dural sac in the ROI; determining one or more parameters associated with the spinal cord or thecal sac in the ROI, wherein the image data of the ROI includes a plurality of axial scans of vertebrae from the plurality of vertebrae, each axial scan of the plurality of axial scans including a cross section of the spinal cord or thecal sac adjacent to the vertebra; and determining a surface area of ​​the cross section of the spinal cord or dural sac in each axial scan from the plurality of axial scans; determining an average surface area of ​​the spinal cord or dural sac based on (1) the surface area of ​​the cross section determined for a first axial scan from the plurality of axial scans corresponding to a most superior scan of the vertebra; and (2) the surface area of ​​the cross section determined for a second axial scan from the plurality of axial scans corresponding to a most inferior scan of the vertebra; identifying one or more surface areas from among remaining surface areas of the cross section determined for the plurality of axial scans excluding the first axial scan and the second axial scan; the one or more surface areas are the smallest or smallest of the remaining surface areas; identifying the one or more surface areas; determining a compression factor based on the one or more surface areas and the average surface area, wherein at least one of the one or more parameters comprises the compression factor; the code further causing the processor to: assessing the severity of spinal stenosis in the ROI based on the one or more parameters associated with the spinal cord or dural sac; A non-transitory processor-readable medium containing said code.

13. 13. The non-transitory processor-readable medium of claim 12, wherein the code for causing the processor to identify the spinal cord or the dural sac comprises code for causing the processor to process the image data of the ROI using a convolutional neural network (CNN) trained to segment the plurality of vertebrae and tissue structures surrounding the plurality of vertebrae to obtain segmentation data that identifies the spinal cord or the dural sac.

14. 13. The non-transitory processor-readable medium of claim 12, wherein the one or more parameters include a cross-sectional area of ​​the spinal cord or dural sac.

15. The non-transitory processor-readable medium of claim 12 , further comprising code for causing the processor to generate a plot of the surface area determined for the cross-sections in the plurality of axial scans.

Images