Methods for aiding in the diagnosis of spinal conditions - Patents.com

Abstract

The present invention relates to a computer-implemented method for providing a diagnostic aid for diagnosing a vertebral condition in a spine of a subject by providing at least one characteristic of the spine of the subject, the method comprising: - a receiving step (S100) receiving at least first and second imaging signals associated with a subject, the at least first imaging signal and / or the second imaging signal being representative of at least a portion of a spine of the subject; - a first processing step (S200), in which a first spine data set is computed based on at least a first imaging signal; - a first prediction step (S300), in which at least one first output relating to a first anatomical structure of the subject's spine is computed based on the first spine data set, the at least first output comprising a first feature of the subject's spine; - a second processing step (S400), in which a second spine data set is computed based on at least the second imaging signals; - a second prediction step (S500), in which at least one second output related to a second anatomical structure of the subject's spine is computed based on the second spine data set and the at least one first output, the at least second output comprising a second feature of the subject's spine representative of at least one spinal condition; Includes.

Description

[Technical field]

[0001] The present invention relates to a computer-implemented method for providing a diagnostic aid for diagnosing a vertebral condition in a spine of a subject by providing at least one characteristic of the spine of the subject.

[0002] The present invention also relates to a system for providing a diagnostic aid for diagnosing a vertebral condition in a spine of a subject by providing at least one characteristic of the spine of the subject to which the above method is performed.

[0003] The present invention relates to the field of medical image analysis, and more particularly to analysis for identifying spinal disorders. [Background technology]

[0004] Low back pain is one of the leading causes of disability worldwide. Low back pain can be due to specific causes (e.g., tumor, infection, or osteoporosis) or nonspecific causes.

[0005] Degenerative disc disease is the most frequent underlying condition associated with non-specific low back pain. For example, the number of patients suffering from lumbar degenerative disc disease is estimated to reach 266 million people worldwide each year.

[0006] Patients with persistent low back pain experience unnecessary long-term pain, anxiety, and reduced quality of life. Furthermore, the economic burden associated with this condition is not negligible. For example, in the United States, costs associated with low back pain have increased from $26.3 billion in 1998 to over $100 billion in 2011.

[0007] In clinical practice, magnetic resonance imaging (MRI) is the gold standard for the diagnosis of low back pain, especially disc degeneration. More precisely, multiple image sequences (e.g., at least T1w and T2w MRI in the sagittal plane, and T2w MRI in the axial plane) are used by clinicians to diagnose one or more of a very wide range of conditions (disc and endplate degeneration, disc herniation, spondylolisthesis, laterolisthesis, central and lateral canal stenosis, vertebral collapse, etc.). This analysis requires both a significant amount of time and spinal expertise that general radiologists do not possess. As a result, the reliability of these diagnoses is moderate to virtually nonexistent.

[0008] It is therefore important to provide tools that allow for earlier diagnosis of back pain in order to improve both patient outcomes and economic and societal costs.

[0009] Platforms have recently been developed with the aim of providing diagnostic aids for spinal conditions based on MRI. More precisely, such platforms are configured to segment acquired MRI images and to output a diagnosis based on measurements performed on computationally processed segmented MRI images.

[0010] However, such a platform is not satisfactory.

[0011] In fact, MRI does not provide a reliable diagnosis of under / overlordosis, scoliosis, and spondylolisthesis according to the standard Meyerding scale. This is because MRI images are acquired without loading the spine, in other words, the spine is in an extended position during the MRI. Therefore, the validity of such platforms to detect the aforementioned conditions is debatable. Moreover, such platforms cannot detect some conditions, such as disc degeneration, endplate degeneration, Schmorl's nodes, vertebral collapse, and even facet arthritis, so clinicians can only partially rely on the output of this platform.

[0012] For example, U.S. Patent No. 62 / 882,076 uses a three-stage approach consisting of: (1) segmentation to allow identification of regions of interest, (2) classification and evaluation of a series of anatomical or pathological measurements, and (3) diagnosis. This approach is inadequate because it relies solely on measurements and does not consider the complexities between multiple pathologies.

[0013] Furthermore, acquired MRIs may be poorly segmented, resulting in unreliable diagnostic output by such platforms.

[0014] It is an object of the present invention to provide an apparatus for determining at least one characteristic of a subject's spine, which is more versatile and reliable than the aforementioned platforms. Summary of the Invention

[0015] A first aspect of the present invention relates to a computer implemented method for providing a diagnostic aid for diagnosing a vertebral condition in a spine of a subject by providing at least one characteristic of the spine of a subject, the method comprising: - a receiving step, where at least first and second imaging signals associated with the object are received, the at least first and / or second imaging signals representing at least a portion of a spine of the object; - a first processing step, in which a first spine data set is computed based on at least the first imaging signal; - a first prediction step, in which at least one first output related to a first anatomical structure of the subject's spine is computed based on the first spine dataset, the at least one first output comprising a first feature of the subject's spine; - a second processing step, in which a second spine data set is computed based on at least the second imaging signals; - a second prediction step, in which at least one second output related to a second anatomical structure of the subject's spine is computed based on the second spine dataset and the at least one first output, the at least second output comprising a second feature of the subject's spine representative of at least one spinal condition; Includes.

[0016] Such a method has many advantages: it allows to make a reliable diagnosis of at least one spinal condition by outputting information, such as first and second features, which take into account: - complexity between different anatomical structures of the spine, where at least a second output related to a second anatomical structure of the subject spine is computed based on the second spine data set and the at least one first output related to the first anatomical structure; and - A complex between several pathologies, where the first and second characteristics may relate to a first and second condition in the spine of a subject.

[0017] According to other advantageous aspects of the invention, the method comprises one or more of the following features, taken alone or in any technically possible combination:

[0018] The first processing step and / or the second processing step include a pre-processing step of computationally processing an image from at least the first and / or second imaging signals, preferably the computationally processed image being successive slices of at least the first and / or second portion of the subject's spine representing a 3D view of at least the first and / or second portion of the subject's spine.

[0019] The pre-processing step also computes at least one of resizing, intensity transformation, artifact correction, bias correction, and / or normalization of pixel intensity distribution on the image.

[0020] The pre-processing step is followed by an anatomical structure segmentation step in which the image is processed to identify boundaries of predetermined anatomical structures.

[0021] The anatomical structure segmentation step is performed by a deep learning model trained on a library of reference spine images, which outputs a set of raw 2D segmentation masks, each mask associated with a corresponding anatomical structure appearing respectively in the preprocessed images, and the deep learning model is preferably a convolutional neural network.

[0022] After the anatomical structure segmentation step, - a post-processing step on each set of raw 2D segmentation masks to remove false positives and artifacts, a labeling step, in which predefined anatomical structures identified in the segmented images are labeled; continues.

[0023] The predetermined anatomical structures include vertebrae and / or intervertebral discs, and for each segmented vertebra in the images acquired after the anatomical segmentation step, a vertebral centroid is calculated, preferably the vertebral centroid is calculated as the center of mass of the segmented vertebra based on the corresponding segmentation mask, and for each intervertebral disc in the images acquired after the anatomical segmentation step, respectively, a disc centroid is calculated, preferably the disc centroid is calculated as the center of mass of the segmented intervertebral disc.

[0024] At least one first output is data or parameters resulting from the calculations performed on the first spine data set, such as grades (numbers), landmarks (vectors), areas (matrix), volumes (tensors), segmented results, positions, probabilities, or weights of an artificial intelligence model, such as a neural network, determined during or after its training to perform the calculations.

[0025] The first feature includes first condition data indicative of an occurrence of a predetermined first condition in the first anatomical structure.

[0026] The first condition is one of: Modic type endplate change, Schmorl's nodes, anterior spondylolisthesis, posterior spondylolisthesis, lateral spondylolisthesis, disc degeneration, hypolordosis, hyperlordosis, scoliosis, disc herniation (symmetric bulging, asymmetric bulging protrusion and prolapse) and its location, separation conditions, nerve root compression conditions, spinal canal stenosis and its origin, lateral recess stenosis and its origin, neural foraminal stenosis and its origin, compression fracture and its acute state, paraspinal muscle atrophy, fatty involution of paraspinal muscles, facet joint disease and its origin, tumor, infection, pain, spondylosis.

[0027] The second feature includes second condition data indicative of an occurrence of a predetermined second condition in the second anatomical structure.

[0028] The occurrence of the second condition is related to the occurrence of the first condition.

[0029] The second anatomical structure is the first anatomical structure.

[0030] The second anatomical structure is separate from the first anatomical structure, and the second anatomical structure is adjacent to the first anatomical structure or located remotely from the first anatomical structure.

[0031] The at least one first output represents a relationship between the first feature and the second feature.

[0032] The data for the first condition indicates the occurrence of Modic-type endplate changes, and the second condition is disc degeneration.

[0033] The first prediction step and the second prediction step are performed by a first and a second neural network, respectively, where the first neural network is trained to generate at least one first output related to a first anatomical structure of the target spine based on a first spine data set, and the second neural network is trained to generate at least one second output related to a second anatomical structure of the target spine based on a second spine data set and the at least one first output, and the at least one first output is related to weights at a particular depth level of the first neural network.

[0034] The first prediction step is configured to predict the presence of a herniated disc and the second prediction step is configured to predict the presence of spinal stenosis; and / or

[0035] The first prediction step is configured to process first features comprising information on the occurrence of a given condition as well as hints for this condition, and the second prediction step is configured to process at least one second output based on the second spine data set and at least one first output comprising the first features and hints.

[0036] Another aspect of the invention relates to a system for providing a diagnostic aid for diagnosing a vertebral condition in a spine of a subject by providing at least one characteristic of the spine of the subject, the system being configured to perform the aforementioned method, the system comprising: a processing device configured to receive at least first and second imaging signals associated with the object, where the at least first and / or second imaging signals represent at least a portion of a spine of the object, and configured to compute first and second spine data sets, respectively, based on the at least first and second imaging signals; a prediction device configured to process at least one first output related to a first anatomical structure of a spine of a subject based on a first spinal dataset, where the at least first output comprises a first feature of the spine of the subject, and configured to process at least one second output related to a second anatomical structure of the spine of the subject based on a second spinal dataset and the at least one first output, where the at least second output comprises a second feature of the spine of the subject indicative of at least one spinal condition; Equipped with.

[0037] The method is designed to be implemented on a remote server, such as a cloud computing platform, that is remotely accessed by a user via a network. In one embodiment, the method includes receiving input data from a user, processing the input data using one or more algorithms to generate an output, and sending the output back to the user via the network. The method may be implemented using any suitable programming language or framework and can take advantage of various types of hardware and software resources available on the remote server. By implementing the method on a cloud computing platform, users can benefit from the scalability, reliability, and cost-effectiveness of cloud-based computing resources while avoiding the need to maintain their own expensive hardware and software infrastructure.

[0038] Another aspect of the present disclosure relates to an apparatus for determining at least one characteristic of a spine of a subject, the apparatus comprising: - at least one first module configured to receive a corresponding first spine image data set associated with the subject, the at least one first module further configured to compute at least one first output related to a first anatomical structure of the examined spine based on the spine image data representative of at least a portion of the examined spine, the at least one first output including a first feature of the examined spine; at least one first output of the first module, and A corresponding second spine image dataset associated with the subject at least one second module, distinct from the first module, configured to receive, based on the spine image data representative of at least a portion of the examined spine and the at least one first output of the first module, at least one second module further configured to compute at least one second output related to a second anatomical structure of the examined spine, the at least one second output including a second feature of the examined spine. The processing unit comprises:

[0039] Indeed, by using the first output of the first module to computationally process the second output of the second module (such as the second feature) for predicting the state, the present disclosure allows taking into account the fact that knowledge of a particular anatomical structure of the spine has an effect either on other anatomical structures, or on the state of the same anatomical structure, due to biomechanical correlations and compensation mechanisms throughout the length of the spine.

[0040] Furthermore, with regard to the processing of imaging signals, by using the first output of the first module for computationally processing the second output of the second module (such as the second feature), the present disclosure makes it possible to combine different acquisition sequences of the same anatomical structure, for example of the spine, to achieve a more reliable segmentation result than would be obtained without implementing the approach according to the present invention.

[0041] According to other advantageous aspects of the present disclosure, the device comprises one or more of the following features, taken alone or in any technically possible combination:

[0042] Each first output received by the second module represents a relationship between a first feature computed by the first module and a second feature computed by the second module.

[0043] The first module is configured to implement a first artificial intelligence model for computing the first feature, and the second module is configured to implement a second artificial intelligence model for computing the second feature, and the first artificial intelligence model and the second artificial intelligence model are configured to provide at least one weight of the first artificial intelligence model that is appropriate for computing the second feature by the second artificial intelligence model during training, and the at least one first output received by the second module includes each weight provided.

[0044] The first feature includes first condition data indicative of an occurrence of a predetermined first condition in a first anatomical structure, and the second feature includes second condition data indicative of an occurrence of a predetermined second condition in a second anatomical structure, the occurrence of the second condition being related to the occurrence of the first condition.

[0045] The second anatomical structure is a first anatomical structure, and occurrence of the second condition in the first anatomical structure is related to occurrence of the first condition in the first anatomical structure.

[0046] At least a portion of the first spine image data and / or the second spine image data represents a shape of the first anatomical structure and / or the second anatomical structure.

[0047] The first spine image data and / or the second spine image data include corresponding labels, corresponding locations, a location of at least one respective landmark, corresponding boundaries, corresponding bounding boxes, and / or corresponding measurements.

[0048] The measurements include distances, areas, volumes, and / or angles within the first anatomical structure and / or the second anatomical structure, and / or distances, areas, volumes, and / or angles between landmarks of the first anatomical structure and / or the second anatomical structure.

[0049] At least one of the first feature and the second feature represents a shape of the first anatomical structure and / or the second anatomical structure.

[0050] At least one of the first feature and the second feature includes a corresponding label, a corresponding location, a location of at least one respective landmark, a corresponding boundary, a corresponding bounding box, and / or a corresponding measurement.

[0051] The measurements include distances, areas, volumes, and / or angles within the first anatomical structure and / or the second anatomical structure, and / or distances, areas, volumes, and / or angles between landmarks of the first anatomical structure and / or the second anatomical structure.

[0052] The first spine image data set includes first data acquired according to a first acquisition sequence, the second spine image data set includes second data acquired according to a second acquisition sequence different from the first acquisition sequence, and the second anatomical structure is the first anatomical structure.

[0053] The first data and the second data include imaging signals representative of the first anatomical structure and / or the second anatomical structure.

[0054] The imaging signal includes an image.

[0055] The present disclosure also relates to a system for detecting at least one condition in a spine of a subject, the system comprising: - a first device as defined above, wherein the first spine image data set and the second spine image data set comprise at least one imaging signal representative of a spine of a subject, the first feature being representative of a shape of a first anatomical structure of the spine of the subject and the second feature being representative of a shape of a second anatomical structure of the spine of the subject, and / or - a second device as defined above, wherein the first spine image data set and the second spine image data set comprise at least one feature representative of at least one shape of a first anatomical structure of the subject's spine and / or a second anatomical structure of the subject's spine, the first feature representative of an occurrence of a first condition in the first anatomical structure and the second feature representative of a second condition in the second anatomical structure. Equipped with.

[0056] According to other advantageous aspects of the present disclosure, the system includes one or more of the following features, taken alone or in any technically possible combination:

[0057] At least a portion of the first spine image data set and the second spine image data set received by the second device are output of the first device.

[0058] The first condition and / or the second condition is a lumbar spine pathology.

[0059] The first condition and / or the second condition is a grade on a predefined scale representing at least one corresponding spinal pathology, such as a lumbar spine pathology, preferably a grade on the Pfirrmann scale and / or a grade on the Modic-type endplate change scale.

[0060] The first spine image data set and / or the second spine image data set of the first device includes at least one of an X-ray radiation imaging signal, a magnetic resonance imaging signal, and an ultrasound signal.

[0061] The present disclosure also relates to a computer-implemented method for determining at least one characteristic of a vertebra of a subject, the method comprising: - providing at least one first artificial intelligence model with a first spine image data set associated with the subject, the first artificial intelligence model being configured to compute at least one first output related to a first anatomical structure of the examined spine based on spine image data representative of at least a portion of the examined spine, the at least one first output comprising a first feature of the examined spine; - at least one second artificial intelligence model different from the first artificial intelligence model, at least one first output of the at least one first artificial intelligence model; and A corresponding second spine image dataset associated with the subject and providing a second spine image data set, wherein the second artificial intelligence model is further configured to compute at least one second output related to a second anatomical structure of the examined spine based on the spine image data representative of at least a portion of the examined spine and the at least one first output of the second artificial intelligence model, wherein the at least one second output includes a second feature of the examined spine. Includes.

[0062] According to other advantageous aspects of the present disclosure, the method comprises one or more of the following features, taken alone or in any technically possible combination:

[0063] Each first output received by the second artificial intelligence model represents a relationship between a first feature computed by the first artificial intelligence model and a second feature computed by the second artificial intelligence model.

[0064] The method includes a step of training a first artificial intelligence model and a second artificial intelligence model to provide at least one weight of the first artificial intelligence model that is appropriate for computationally processing a second feature by the second artificial intelligence model, and at least one first output received by the second artificial intelligence model includes each provided weight.

[0065] At least one of the first feature and the second feature represents a shape of the first anatomical structure and / or the second anatomical structure.

[0066] At least one of the first feature and the second feature includes a corresponding label, a corresponding location, a location of at least one respective landmark, a corresponding boundary, a corresponding bounding box, and / or a corresponding measurement.

[0067] The measurements include distances, areas, volumes, and / or angles within the first anatomical structure and / or the second anatomical structure, and / or distances, areas, volumes, and / or angles between landmarks of the first anatomical structure and / or the second anatomical structure.

[0068] The first spine image data set includes first data acquired according to a first acquisition sequence, the second spine image data set includes second data acquired according to a second acquisition sequence different from the first acquisition sequence, and the second anatomical structure is the first anatomical structure.

[0069] The first data and the second data include imaging signals representative of the first anatomical structure and / or the second anatomical structure.

[0070] The imaging signal includes an image.

[0071] According to further advantageous aspects of the present disclosure, the method may comprise one or more of the following features, alone or in any technically possible combination (and / or in combination with the aforementioned features):

[0072] The first feature includes first condition data indicative of an occurrence of a predetermined first condition in a first anatomical structure, and the second feature includes second condition data indicative of an occurrence of a predetermined second condition in a second anatomical structure, the occurrence of the second condition being related to the occurrence of the first condition.

[0073] The second anatomical structure is a first anatomical structure, and occurrence of the second condition in the first anatomical structure is related to occurrence of the first condition in the first anatomical structure.

[0074] At least a portion of the first spine image data and / or the second spine image data represents a shape of the first anatomical structure and / or the second anatomical structure.

[0075] The first spine image data and / or the second spine image data include corresponding labels, corresponding locations, a location of at least one respective landmark, corresponding boundaries, corresponding bounding boxes, and / or corresponding measurements.

[0076] The measurements include distances, areas, volumes, and / or angles within the first anatomical structure and / or the second anatomical structure, and / or distances, areas, volumes, and / or angles between landmarks of the first anatomical structure and / or the second anatomical structure.

[0077] The present disclosure also relates to a computer program product comprising instructions that, when executed by a computer, cause the computer to automatically perform the steps of a method for providing a diagnostic aid for diagnosing a vertebral condition in a subject's spine by providing at least one characteristic of the subject's spine according to any one of the above-mentioned embodiments (i.e., the instructions may cause the computer to perform the steps of the method according to one of those embodiments, or according to two or more of those embodiments). The computer program product may be implemented on a remote server, such as a cloud computing platform, remotely accessed by a user via a network.

[0078] The present disclosure also relates to a non-transitory computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the steps of the method according to any one of the above embodiments.

[0079] Such non-transitory program storage devices may be, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor devices, or any suitable combination of the foregoing. The following provides more specific examples, but should be understood as merely illustrative and not exhaustive, as can be readily understood by those skilled in the art: portable computer diskettes, hard disks, ROM, EPROM (erasable programmable ROM) or flash memory, portable CD-ROM (compact disc ROM). Furthermore, the non-transitory computer-readable storage medium may be implemented on a remote server, such as a cloud computing platform, which is accessed remotely by a user via a network.

[0080] The invention will be better understood with reference to the accompanying drawings. [Brief description of the drawings]

[0081] [Figure 1]1 is a schematic diagram of a first embodiment of a computer-implemented method for providing assistance in diagnosing a vertebral condition in a spine in a subject; [Diagram 2] 1 is a schematic diagram of a first embodiment of a spinal condition detection system according to the present disclosure. [Diagram 3] 3 is an example showing the operation of the spinal condition detection system of FIG. 2. [Figure 4] 1 is a schematic diagram of a second embodiment of a spinal condition detection system according to the present disclosure. [Diagram 5] 1 is a schematic diagram of a third embodiment of a spinal condition detection system according to the present disclosure.

[0082] definition According to the present invention, the expression "anatomical structure" refers to a biological structure of the spinal region that is distinct from adjacent structures. The anatomical structure may include vertebrae, ligaments, tendons, spinal cord, nerve roots, pedicles, neural foramina, intervertebral discs, facet joints, facet joints (or synovial joints), joint capsules, paraspinal muscles, spinal segments, lumbar, thoracic, cervical vertebrae, or parts thereof.

[0083] According to the present invention, the expression "shape of an anatomical structure" refers to a corresponding label, a corresponding position, a position of at least one respective landmark, a corresponding boundary, a corresponding bounding box, and / or corresponding measurements, the measurements preferably including distances, areas, volumes, and / or angles within the anatomical structure and / or distances, areas, volumes, and / or angles between landmarks of the anatomical structure.

[0084] According to the present invention, the expression "imaging signal" refers to a signal output by a medical imaging device and representing at least one imaged anatomical structure. The imaging signal may include a time domain signal, a spectral domain signal, and / or an image of at least one imaged anatomical structure. The imaging signal may also include clinical data about the examined human (e.g., age, sex, size, weight, activity, past or present medical condition(s), etc.) and information about the medical imaging device (e.g., focal length, field of view size, resolution, contrast, orientation, imaged anatomical plane, etc.).

[0085] According to the present invention, the term "condition" refers to a condition of the spine, including one or more anatomical structures (s) of the spine, either adjacent or non-adjacent. The condition may be a spinal pathology, disease, and / or injury.

[0086] According to the present invention, the expression "processing unit" refers to a processing device capable of executing instructions for performing associated and / or resulting functions, regardless of their implementation.

[0087] According to the present invention, the expression "spine image data" refers to data determined on the basis of imaging at least a part of a spine of a subject. Spine image data may include imaging signals (as defined above) and / or information computed on the basis of said imaging signals. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0088] The present description illustrates the principles of the disclosure and it will thus be understood that those skilled in the art may devise various arrangements which, although not explicitly described or shown herein, embody the principles of the disclosure and are included within its spirit and scope.

[0089] All examples and conditional language cited herein are for educational purposes to aid the reader in understanding the principles of the present disclosure and the concepts the inventors have contributed to the development of the art, and should be construed as not being limited to such specifically cited examples and conditions.

[0090] Moreover, all statements herein reciting principles, aspects, and specific embodiments of the disclosure, are intended to encompass both structural and functional equivalents thereof. Additionally, such equivalents are intended to include both currently known equivalents as well as equivalents developed in the future, i.e., any elements developed that perform the same function, regardless of structure.

[0091] Thus, for example, those skilled in the art will appreciate that the block diagrams presented herein may represent conceptual views of illustrative circuitry embodying the principles of the present disclosure. Similarly, it will be appreciated that any flow charts, flow diagrams, or the like may be substantially represented in a computer-readable storage medium and represent various processes performed by a computer or processor, whether or not such a computer or processor is explicitly shown.

[0092] The functions of the various elements shown in the figures may be provided through the use of dedicated hardware as well as hardware capable of executing software in association with appropriate software. When provided by a processor, the functions may be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some of which may be shared.

[0093] It should be understood that the elements shown in the figures may be implemented in various forms of hardware, software, or combinations thereof. Preferably, these elements are implemented in a combination of hardware and software on one or more appropriately programmed general-purpose devices, which may include a processor, memory, and input / output interfaces.

[0094] First Method Embodiment FIG. 1 illustrates a first embodiment of a computer-implemented method for providing a diagnostic aid for diagnosing a vertebral condition in a spine of a subject by providing at least one characteristic of the spine of the subject.

[0095] The method comprises: - a receiving step S100, in which at least first and second imaging signals associated with a subject are received, the at least first and / or second imaging signals representing at least a portion of a spine of the subject; a first processing step S200, in which a first spine data set is computed based on at least a first imaging signal; - a first prediction step S300, in which at least one first output relating to a first anatomical structure of the subject spine is computed based on the first spine data set, the at least one first output comprising a first feature of the subject spine; a second processing step S400, in which a second spine data set is computed based on at least the second imaging signal; - a second prediction step S500, in which at least one second output related to a second anatomical structure of the subject's spine is computed based on the second spine data set and the at least one first output, the at least second output comprising a second feature of the subject's spine representative of at least one spinal condition; Includes.

[0096] Advantageously, the method may also include a generating step S600, in which a medical report is generated to assist the physician attending to the subject in diagnosing the presence of at least one condition in the subject's spine. If there is at least one condition detected or not detected, the medical report assists the subject in explaining what condition it is or is not and where it is.

[0097] The medical report may be fully customizable. For example, the medical report may include some of the images processed during the first processing step S200 and the second processing step S400 to indicate the location of the condition and the measurements performed to detect the condition, as well as information related to the first prediction step S300 and the second prediction step S500, such as the first and second features.

[0098] Reception step S100 Advantageously, at least the first and / or second imaging signals may include images from medical imaging devices, such as, for example, X-ray or ultrasound devices, MRI or CT scans, but not limited to this list. In some cases, at least the first and / or second imaging signals may include images of the same medical imaging device, having the same modality (e.g., T1, T1w, T2, T2w, STIR, FLAIR, 3D, Dixon, etc.) or different modalities, acquired during the same imaging sequence or acquired from different sequences. In other cases, the first and / or second imaging signals may include images of different medical devices. For example, at least the first and / or second imaging signals may include images with a combination of different modalities (e.g., T1w and T2w, X-ray and ultrasound, etc.), and with the same anatomical plane or different anatomical planes between sagittal, axial, coronal planes.

[0099] Preferably, the images obtained from the medical imaging device are 2D images, but may also be 3D images comprising successive slices of the 2D images, particularly for the diagnosis of detailed medical conditions.

[0100] Additionally, at least the first and / or second imaging signals may include clinical data with or without images from the medical imaging device(s), which may relate to the subject under examination (e.g., age, sex, size, weight, activity, past or present medical condition(s), etc.) as well as information about the medical imaging device(s) used to acquire the at least first and second imaging signals (focal length, field of view size, resolution, contrast, orientation, anatomical plane imaged, etc.).

[0101] At least the first and second imaging signals may also be identical. Advantageously, in this case, some processing may be interconnected, which makes it possible to reduce processing times, in particular during the first processing step S200 and the second processing step S400.

[0102] Thus, compared to conventional methods that use only one type of view or only one modality, the present method can diagnose spinal conditions based on a combination of views (sagittal, axial, and / or coronal), modalities (such as T1, T1w, T2, T2w, STIR, FLAIR, 3D, and / or Dixon), and clinical data, allowing for a more accurate and reliable diagnosis.

[0103] First Processing Step S200 and Second Processing Step S400 If at least the first and second imaging signals are identical, the first and second spine data sets computed by the first processing step S200 and the second processing step S400 may be identical and / or may have inter-linked sub-steps.

[0104] Regardless of whether the first and second spine data sets are identical or not, the steps of computing the first and second spine data sets in the first processing step S200 and / or the second processing step S400 may include one or more sub-steps, which may be applied individually, in combination or in succession, such as a pre-processing step, anatomic structure segmentation, a post-processing step, a center of gravity computation, a disc position computation, a labeling step, etc.

[0105] The pre-processing step computes an image from at least the first and / or second imaging signals, preferably the computed image being successive slices of at least the first and / or second portion of the subject's spine and representing a 3D view of at least the first and / or second portion of the subject's spine.

[0106] Additionally, the pre-processing step may also compute at least one of resizing, intensity transformation, artifact correction, bias correction, and / or normalization of pixel intensity distribution on the image.

[0107] An anatomical structure segmentation step is typically performed on the pre-processed images to identify boundaries of anatomical structures of interest, including vertebrae, ligaments, tendons, spinal cord, nerve roots, pedicles, neural foramina, intervertebral discs, facet joints, facet joints (or synovial joints), joint capsules, paraspinal muscles, spinal segments, lumbar, thoracic, cervical vertebrae, or portions thereof.

[0108] The anatomy segmentation step may be performed by a deep learning model, such as a convolutional neural network, which is trained on a library of reference spine images where the given anatomy has been outlined by a medical practitioner or the segmented images have been verified by a medical practitioner during training.

[0109] Advantageously, a deep learning model can be a collection of different models (which may or may not have been trained on the same training data) that have different architectures, or models with similar architectures but different weights. The segmentations made by the ensemble models can then be combined through voting statistics or more sophisticated methods that learn how much to trust each model and under what conditions.

[0110] Ensemble models have the advantage over single models of improving performance and segmentation, and have additional robustness due to the spread of segmentation and the voting of each model in the ensemble.

[0111] For example, the intervertebral discs can be directly identified by segmentation and object detection models, but can also be extrapolated from vertebral segmentation. Each of these approaches has advantages and disadvantages depending on the case. Combining all these pipelines ensures a more robust and high-performance model in real applications.

[0112] As mentioned above, the method, and therefore the deep learning model, may allow different views and modalities to be analyzed. For example, but not limited to, the deep learning model may be trained on the following types of images: sagittal T1w MRI sequences in DICOM format, sagittal T2w MRI sequences in DICOM format, sagittal T2 STIR MRI sequences in DICOM format, axial T2w MRI sequences in DICOM format, coronal T1w MRI sequences in DICOM format, coronal T2 STIR MRI sequences in DICOM format, sagittal DIXON MRI sequences in DICOM format, and / or 3D T2w MRI sequences in DICOM format.

[0113] Advantageously, the deep learning model, preferably a convolutional neural network, outputs a set of raw 2D segmentation masks, each mask associated with a corresponding anatomical structure respectively appearing in the pre-processed image(s).

[0114] Thus, the output of the anatomy segmentation step is a crop of the region of interest, ie the disc and the area surrounding the vertebrae, and the vertebrae may be labeled, ie the anatomical level of each vertebra.

[0115] A post-segmentation processing step may be applied to each set of raw 2D segmentation masks to remove false positives and artifacts. An effective way to do this is to compare successive segmented images acquired after the anatomical structure segmentation step, where at least the first and / or second imaging signals comprise successive slices of at least the first and / or second portions of the spine of interest. This allows for identifying common structures appearing in multiple slices and removing noise / artifacts present in the 2D slices.

[0116] If the given anatomical structure includes vertebrae and / or intervertebral discs, for each segmented vertebra in the image acquired after the anatomical segmentation step, a vertebral centroid and / or intervertebral disc centroid may be computed. Preferably, the vertebral centroid may be computed as the center of mass of the segmented vertebra identified in the segmented image, such as in the corresponding segmentation mask. Respectively, the intervertebral disc centroid may be computed as the center of mass of the segmented disc identified in the segmented image.

[0117] Another way to determine the centroid of the intervertebral disc is to perform a polynomial regression of the spinal curve using the centroids of the vertebrae as anchor points. The degree of the polynomial can be defined as the number of anchor points identified. This allows for the interpolation of the centroid of the intervertebral disc (IVD) as a point on the spinal curve that is the mid-distance between the centroids of the two vertebrae.

[0118] The labeling step consists of labeling certain anatomical structures identified in the images acquired after the anatomical structure segmentation step. The labeling step is usually applied after a post-processing step of segmentation for better labeling, but it may also be applied at the end of the anatomical structure segmentation step. For example, the computed centroid of each vertebra is assigned to the corresponding vertebra, and the computed centroid of each intervertebral disc is assigned to the corresponding intervertebral disc.

[0119] Furthermore, bounding boxes may be computed on the segmented images and around the regions / anatomical structures of interest, such as the intervertebral discs. The method to compute bounding boxes around the intervertebral discs is to rotate each labeled mask with respect to the spinal curvature angle, and for each intervertebral disc of interest, consider the surrounding vertebrae and rotate their centers with respect to the spinal curvature angle as well. The bounding box of the intervertebral disc is then computed by clipping the rotated labeled mask by using the rotated center coordinates and offsets of the surrounding vertebrae. The distance between the intervertebral disc and its adjacent vertebrae may be used to compute the height of the bounding box, so that each bounding box represents a region of interest that includes the above anatomical structures. The same process can be performed to compute bounding boxes around each vertebra.

[0120] First Prediction Step S300 and Second Prediction Step S500 Advantageously, at least one first output computed in the first prediction step S300 is data or parameter(s) of the result of the computation performed on the first spine data set, such as grade (numerical value), landmark (vector), area (matrix), volume (tensor), segmented result, position, probability or weight of an artificial intelligence model, such as a neural network, determined during or after training to perform the computation.

[0121] The first feature may include first condition data indicative of an occurrence of a predetermined first condition in the first anatomical structure.

[0122] For example, the first condition can be one of Modic-type endplate changes, Schmorl's nodes, anterior spondylolisthesis, posterior spondylolisthesis, lateral spondylolisthesis, disc degeneration, hypolordosis, hyperlordosis, scoliosis, disc herniation (symmetric bulging, asymmetric bulging protrusion and prolapse) and its location, a separation condition, a nerve root compression condition, spinal stenosis and its origin, lateral recess stenosis and its origin, neural foraminal stenosis and its origin, a compression fracture and its acute condition, paraspinal muscle atrophy, fatty involution of paraspinal muscles, facet joint disease and its origin, a tumor, an infection, pain, and spondylosis.

[0123] Similarly, the second feature may include second condition data indicative of the occurrence of a predetermined second condition in the second anatomical structure.

[0124] Advantageously, the occurrence of the second condition may be associated with the occurrence of the first condition.

[0125] In some cases, the second anatomical structure is the first anatomical structure, and in other cases, the second anatomical structure may be different from the first anatomical structure, the second anatomical structure being adjacent to the first anatomical structure or located remotely from the first anatomical structure.

[0126] Furthermore, the at least one first output may advantageously represent a relationship between the first feature and the second feature.

[0127] For example, data from a first condition may indicate the occurrence of Modic-type endplate changes and data from a second condition may indicate the occurrence of disc degeneration.

[0128] Thus, the output of the first prediction step S300 and the second prediction step S500 may be, for each anatomical level present in the image and detected by the previous processing steps S200 and S400, an identification of the pathologies present and an assessment of their grade according to criteria specific to each pathology.

[0129] Advantageously, the first prediction step S300 and the second prediction step S500 are respectively performed by a first and a second neural network, the first neural network being trained to generate at least one first output related to a first anatomical structure of the target spine based on the first spine data set and the second neural network being trained to generate at least one second output related to a second anatomical structure of the target spine based on the second spine data set and the at least one first output, the at least one first output being related to weights at a particular depth level of the first neural network.

[0130] The training sets for the first and second neural networks may be the same training sets used for the deep learning model, or may be one or more image libraries in which a physician has already identified pathologies / conditions in the imaged spine.

[0131] In another example, the first prediction step may be configured to predict the presence of a herniated disc and the second prediction step may be configured to predict the presence of spinal stenosis.

[0132] Advantageously, the first prediction step is configured to process a first feature comprising information on the occurrence of a given condition as well as an indication of this condition, and the second prediction step is configured to process at least one second output based on the second spine data set and at least one first output comprising the first feature and the indication.

[0133] Table 1 shows examples of the types of imaging signals that may be used and processed to predict several disease states / conditions according to the methods of the present invention, and possible relationships and correlations between various disease states and their predictions.

[0134] [Table 1]

[0135] For example, if the diagnosis of several conditions, such as Pfirmann, Modic, vertebral fracture, vertebral tray / plate, use the same type of image as input, the first processing step S200 and the second processing step S400 may be combined together to reduce processing time.

[0136] Furthermore, the prediction of a first state in this list by the first prediction step S300 may improve the probability and reliability of predicting another state that is correlated with the first state during the second prediction step S500.

[0137] The computer program implementing the method of the present embodiment can be typically distributed to users via a computer-readable storage medium for distribution, such as, but not limited to, an SD card, an external storage device, a microchip, a flash memory device, a portable hard drive, a software website, etc. From the distribution medium, the computer program can be copied to a hard disk or a similar intermediate storage medium. The computer program can be executed by loading the computer instructions from the distribution medium or intermediate storage medium into the computer's execution memory and configuring the computer to operate according to the method of the present disclosure. All these operations are well known to those skilled in the art of computer systems.

[0138] The instructions or software for controlling a processor or computer to implement the hardware components and perform the above methods, and any associated data, data files, and data structures, are recorded, stored, or fixed in or on one or more non-transitory computer-readable storage media. Examples of non-transitory computer-readable storage media include read-only memory (ROM), random-access memory (RAM), flash memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-Res, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state disk, and any device known to one of skill in the art that can store the instructions or software and any associated data, data files, and data structures in a non-transitory manner and can provide the instructions or software and any associated data, data files, and data structures to a processor or computer, thereby allowing the processor or computer to execute the instructions. In one example, the instructions or software and any associated data, data files, and data structures are distributed across a network connected computer system such that the instructions and software and any associated data, data files, and data structures are stored, accessed, and executed by the processors or computers in a distributed format.

[0139] First System Embodiment A system 2A according to a first system embodiment of the present disclosure is shown in Figure 2. System 2A provides diagnostic aids for diagnosing spinal conditions and is configured to implement the methods described above.

[0140] The system 2A includes an image processing device 4A and a prediction device 6A.

[0141] The image processing device 4A (also referred to as "image processing unit") is configured to process at least one imaging signal associated with the subject, more specifically one or more imaging signal(s) representative of at least a portion of the subject's spine, such as one or more images of at least a portion of the subject's spine. The image processing device 4 is further configured to output spine image data associated with the subject and computationally processed based on the at least one imaging signal. Thus, the image processing device 4A is configured to perform the receiving step S100 and the first and second processing steps S200 and S400 as described above.

[0142] Furthermore, the prediction device 6A (also referred to as "prediction unit") is configured to compute at least one feature of the target spine based on the spine image data output by the image processing device 4A. For example, the at least one feature includes status data indicative of the occurrence of at least one predetermined status for the target spine. Thus, the prediction device 6A is configured to perform the first prediction step S300 and the second prediction step S500 as described above.

[0143] According to the present disclosure, the terms "image processing unit" and "prediction unit" should not be construed as being limited to hardware capable of executing software, but generally refer to a processing device that may comprise, for example, a microprocessor, an integrated circuit, or a programmable logic device (PLD). Each of the image processing unit and the prediction unit may include one or more graphics processing units (GPUs) or tensor processing units (TSUs), whether utilized for computer graphics and image processing or other functions. Furthermore, the terms "image processing unit" and "prediction unit" should not be construed as being limited to separate processing devices. Furthermore, instructions and / or data enabling the implementation of the relevant functions and / or resulting functions may be stored in any processor-readable medium, such as, for example, an integrated circuit, a hard disk, an optical disk such as a CD (compact disk), a DVD (digital versatile disk), a RAM (random access memory), or a ROM (read only memory). The instructions may in particular be stored in hardware, software, firmware, or any combination thereof.

[0144] Image Processing Unit 4A As mentioned above, the image processing unit 4A is configured to process at least one input imaging signal associated with the subject in order to compute associated spine image data.

[0145] The respective imaging signals may be, for example, X-ray imaging signals, Magnetic Resonance Imaging (MRI) signals, or ultrasound signals, such as X-ray imaging images, Magnetic Resonance Imaging (MRI) images, or ultrasound images.

[0146] In each of the three imaging techniques described above, imaging signals may be acquired while the subject is in a stationary position (static acquisition) or while the subject is in multiple successive static positions, e.g., flexion followed by extension (also referred to as "dynamic acquisition").

[0147] Furthermore, in each of the three imaging modalities discussed above, the imaging signals may be acquired after injecting a contrast agent into the subject.

[0148] For example, the x-ray imaging signals are acquired using at least one of the following techniques: computed tomography, real-time radiography, dual-energy x-ray absorptiometry (DEXA), and / or low-dose biplanar whole-body radiography in functional position (also referred to by its trade name "EOS").

[0149] For example, ultrasound imaging signals may be acquired using standard ultrasonography and / or ultrasound elastography, which is well suited for determining mechanical properties of the imaged tissue.

[0150] For example, MRI imaging signals may be acquired using standard MRI and / or magnetic resonance elastography, which is well suited for determining the mechanical properties of the imaged tissue.

[0151] Preferably, the image processing unit 4A is configured to process imaging signals (more specifically 2D images) acquired according to different acquisition techniques, for example in the case of MRI imaging signals the image processing unit 4A is configured to process imaging signals acquired according to different imaging sequences (T1w, T2wSTIR, FLAIR, 3D, Dixon, etc.) in at least one of the three anatomical planes (sagittal, axial, coronal).

[0152] Hereinafter, the acquired imaging signal is also referred to as the "input imaging signal."

[0153] The image processing unit 4A is configured to perform computational processing of image data, for example by performing a pre-processing step, an anatomical structure segmentation step, a post-segmentation processing step, a centroid calculation processing step, and an intervertebral disc position calculation processing step on the input imaging signal.

[0154] Pretreatment As mentioned above, the image processing unit 4A is preferably configured to apply at least one pre-defined pre-processing transformation to each input imaging signal during the pre-processing step.

[0155] Furthermore, if the input imaging signal is not an image, the image processing unit 4A is configured to perform computational processing of an image on the or each input imaging signal during a pre-processing step.

[0156] For example, image processing unit 4A is configured to apply at least one of resizing, intensity transformation (such as intensity normalization or intensity clipping), artifact correction (such as bias field correction), bias correction, and pixel intensity distribution normalization (such as histogram equalization or registration methods) during the pre-processing step.

[0157] In the case of pixel intensity distribution normalization, the image processing unit 4A is configured to modify the pixel intensity distribution of the input image, for example, to match a reference pixel intensity distribution, such a reference pixel intensity distribution being preferably a pixel intensity distribution of an image used for training at least one artificial intelligence model implemented by the prediction device 6A, as described below.

[0158] Segmentation of anatomical structures The image processing unit 4A is also configured to apply segmentation to each image (the preprocessed image if a preprocessing step is performed, or the input image if a preprocessing step is not performed) during the anatomical structure segmentation step in order to identify boundaries of predetermined anatomical structures.

[0159] Such anatomical structures include, for example, vertebrae, intervertebral discs, paraspinal muscles, facet joints, nerves, cysts, bone spurs, edema, and / or blood vessels (particularly arteries and vena cava).

[0160] For example, the image processing unit 4A is configured to implement a deep learning model, preferably a 2D convolutional neural network (such as a neural network with a U-Net architecture), to perform such segmentation. In this case, during the anatomical structure segmentation step, each input image (or preprocessed image) is provided to a deep learning model, preferably a conventional neural network, and for each image, a set of raw 2D segmentation masks is obtained as output. Each mask is associated with a corresponding anatomical structure (also referred to as a "segmented anatomical structure") appearing on the image.

[0161] An image (raw or post-processed) associated with a corresponding segmentation mask is also called a "segmented image."

[0162] Post-segmentation processing The image processing unit 4A is also preferably configured to apply at least one predefined post-processing transformation to each set of raw 2D segmentation masks during a post-processing step, such a step aiming at removing, for example, false positives and artifacts.

[0163] For example, the image processing unit 4A is configured to perform morphological and mathematical operations (erosion, dilation, closing, opening, watershed transformation, etc.) and connected components methods on the raw 2D segmentation mask or on its binarized version during a post-processing step, which makes it possible to remove connected components of the raw 2D segmentation mask that are, for example, smaller than a pre-set threshold or, in the case of a mask corresponding to a vertebra, too far from the rest of the vertebra.

[0164] For input images corresponding to successive slices of at least a portion of the target spine, representing a 3D view of at least a first and / or second portion of the target spine, the image processing unit 4A is further configured to compare successive segmented images as a whole, for example by implementing a voting algorithm, which allows to identify anatomical structures appearing in the segmented images corresponding to successive slices and to remove noise and / or artifacts present in the segmented images.

[0165] Center of gravity calculation Preferably, when the image processing unit 4A is configured to determine the vertebra segmentation mask, the image processing unit 4A is also configured to compute the vertebral centroid of each segmented vertebra. In this case, for each vertebra, the image processing unit 4A is preferably configured to compute the vertebral centroid as the center of mass of the segmented vertebra based on the corresponding segmentation mask. In other words, to compute the centroid of a particular vertebra, the image processing unit 4A is configured to take into account, for each image representing a vertebra, the segmentation mask corresponding to that vertebra.

[0166] For example, for a particular vertebra, the image processing unit 4A is configured to determine a number of partial centroids. More precisely, each partial centroid is the geometric center of a section of the vertebra on the respective 2D image representing the vertebra. In this case, the centroid of a particular vertebra is the center of mass of all partial centroids corresponding to that vertebra.

[0167] The image processing unit 4A is further configured to assign the computed centroid of each vertebra to a corresponding vertebra.

[0168] The image processing unit 4A is advantageously configured to apply similar processing to determine the centroids of other segmented anatomical structures of the spine region.

[0169] Preferably, the center of gravity of the vertebra that is lowest along the longitudinal axis of the spine is identified as the center of gravity of the sacrum S1. Alternatively, the centers of gravity of the vertebrae S2-S5 may also be identified.

[0170] Calculation of intervertebral disc position The image processing unit 4A may be configured to determine the location of the centroid of the intervertebral disc, in which case the image processing unit 4A is configured to determine as the centroid of the intervertebral disc a point on the spinal curve at a mid-distance between the centroids of two consecutive vertebrae.

[0171] More precisely, the image processing unit 4A is preferably configured to determine the aforementioned spinal curve. To that end, the image processing unit is advantageously configured to perform a polynomial regression, for example using the centroids of the vertebrae as anchor points. The image processing 4A may further be configured to select the degree of the polynomial as the number of anchor points.

[0172] Alternatively, instead of interpolating the disc centroids from the segmented vertebrae, the image processing unit 4A is configured to perform segmentation directly on the discs, in which case the image processing unit 4A is configured to assign, for each disc, the corresponding disc centroid to the centre of mass of the disc.

[0173] The image processing unit 4A is preferably further configured to compute a bounding box around the centroid of each disc, for example by computing the height of the bounding box using the distance between the disc and its adjacent vertebrae.More generally, for each identified anatomical structure, the corresponding bounding box may be considered as a region of interest containing that anatomical structure.

[0174] Preferably, the image processing unit 4A is further configured to perform, for each intervertebral centroid, a rotation of the corresponding bounding box using the normal of the spinal curve at the intervertebral centroid. As a result, the bounding boxes all have the same orientation from one segmented image to another, thereby enabling the system 2A to be invariant to the relative rotation of the intervertebral disc from one segmented image to another. More generally, for the bounding boxes, the image processing unit 4A may be configured to perform a similar rotation such that for a particular anatomical structure, the corresponding bounding boxes have the same orientation from one segmented image to another segmented image.

[0175] Alternatively, the image processing unit is configured to perform an object detection method to directly determine bounding boxes and labels of vertebrae, intervertebral discs, and other identified anatomical structures. For example, to determine a bounding box around an intervertebral disc, the image processing unit 4A is configured to implement a mask region-based deep learning model, preferably a Convolutional Neural Network (R-CNN).

[0176] The spine image data output by image processing unit 4A and received by prediction unit 6A includes, for each anatomical structure, the corresponding determined position (or the position of each landmark), the corresponding boundary (i.e. the outer limit), the corresponding bounding box(es), associated measurements (distances, areas, volumes, angles, etc. within the anatomical structure and between its landmarks), and / or the corresponding label(s). Preferably, the spine image data also includes each image and the corresponding segmentation mask.

[0177] For example, in the case of vertebrae and intervertebral discs, the spine image data computed by image processing unit 4A may include the location of the centroid of each vertebra, the label of each vertebra, the location of the centroid of each disc, associated measurements (width, height, orientation, etc.), and / or the label of each disc, as well as each acquired imaging signal (e.g., each acquired image) associated with a corresponding segmentation mask.

[0178] Prediction Unit 6A As mentioned above, prediction unit 6A is configured to compute at least one feature of the target spine based on the spine image data provided by image processing unit 4A, each feature including, for example, condition data indicative of the occurrence of at least one predetermined condition with respect to the target spine.

[0179] The prediction unit 6A comprises at least a first module 8 (also referred to as the "first prediction module") and a second module 10 (also referred to as the "second prediction module"). In the example of Fig. 2, the prediction unit 6A further comprises additional modules 12 (also referred to as the "third prediction module") and 14 (also referred to as the "fourth prediction module").

[0180] First prediction module 8 The first prediction module 8 is configured to receive a first data set of spine image data associated with the subject from the spine image data output by the image processing unit 4A.

[0181] The first prediction module 8 is also configured to compute at least one first output related to a first anatomical structure of the examined spine based on the spine image data representative of at least a portion of the examined spine. Thus, the first prediction module 8 is configured to compute at least one first output related to a first anatomical structure of the current patient's spine based on the first spine image dataset, where the at least one first output includes a first feature of the examined spine.

[0182] For example, to compute each first output, the first prediction module 8 is configured to implement an artificial intelligence model.

[0183] Furthermore, according to the present disclosure, the "first output of the first module" means any data or parameter (weight, etc.) that is the result of the calculation performed by the first prediction module 8. Such data can be, for example, a grade (numerical value), a landmark (vector), an area (matrix), a segmentation result, a position, a probability, etc. Furthermore, each parameter can be a weight of an artificial intelligence model, determined either during or after the training of this model.

[0184] The first feature preferably comprises first condition data (included in said condition data) indicative of the occurrence of a predetermined first condition in the first anatomical structure.

[0185] The first condition (and the second condition described below) may be, for example, one of Modic endplate changes, Schmorl's nodes, anterior spondylolisthesis, posterior spondylolisthesis, lateral spondylolisthesis, disc degeneration, hypolordosis, hyperlordosis, scoliosis, disc herniation (symmetric bulging, asymmetric bulging protrusion and prolapse) and its location, separation conditions, nerve root compression conditions, spinal stenosis and its origin, lateral recess stenosis and its origin, neural foraminal stenosis and its origin, compression fractures and their acute conditions, paraspinal muscle atrophy, fatty involution of paraspinal muscles, facet joint disease and its origin, tumor, infection, pain, spondylosis, and the like.

[0186] Second Prediction Module 10 The second prediction module 10 is configured to receive a second spine image data set associated with the subject from the spine image data output by the image processing unit 4A. The second spine image data set may be the same as or different from the first spine image data set.

[0187] The second prediction module 10 is also configured to receive at least one first output of the first prediction module 8. This is indicated in FIG. 1 by an arrow pointing from the first prediction module 8 to the second prediction module 10.

[0188] The second prediction module 10 is further configured to compute at least one second output related to a second anatomical structure of the examined spine based on the spine image data representative of at least a portion of the examined spine and the at least one first output of the first module. Thus, the second prediction module 10 is configured to compute at least one second output related to a second anatomical structure of the current patient's spine based on the at least one first output and the second spine image dataset. In this case, the at least one second output comprises a second feature of the examined spine.

[0189] Preferably, the second feature comprises data of a second condition indicative of the occurrence of a predetermined second condition in the second anatomical structure. More precisely, the occurrence of the second condition is related to the occurrence of the first condition. In this case, the second anatomical structure is preferably the first anatomical structure or a second anatomical structure of the spine that is separate (i.e., different) from the first anatomical structure, for example an anatomical structure adjacent to the first anatomical structure or an anatomical structure located at a distance from the first anatomical structure.

[0190] For example, to compute each second output, the second prediction module 10 is configured to implement an artificial intelligence model.

[0191] As mentioned above, each first output of the first prediction module 8 input to the second prediction module 10 may be data or parameters (such as weights) computed by the first prediction module 8. The determination of the data may be determined after or during training of the first and second prediction modules 8, 10.

[0192] For example, the second prediction module 10 is configured to compute each second output using at least some of the weights of the model implemented by the first prediction module 8 .

[0193] Each first output of the first prediction module 8 that is input to the second prediction module 10 may be selected using a qualitative approach or a quantitative approach, or a combination of both approaches.

[0194] For example, each first output input to the second prediction module 10 may be quantitatively selected during configuration (i.e., training) of the first and second prediction modules 8, 10 by implementing an algorithm (attention mechanism, gradient descent, etc.) adapted to highlight weight patterns of the model of the first prediction module 8 relative to the model of the second prediction module 10.

[0195] As a first example, a first prediction module 8 configured to predict the occurrence of Schmorl's nodes and a second prediction module 10 configured to predict the occurrence of Modic-type endplate changes are considered. Furthermore, the model is trained to determine co-attention coefficients between layer I of the first module 8 and layer j of the second module 10. These co-attention coefficients represent which portion of the high-level features of layer i of the first prediction module 8 are input to layer j of the second prediction module 10. This configuration is advantageous because Schmorl's nodes and Modic-type endplate degeneration share similar indicators such as endplate localization and hyperintensity in T2w images. Thus, mid-level and high-level features can be easily shared via co-attention connections between the Schmorl's nodes prediction module (i.e., the first prediction module 8) and the Modic-type endplate changes prediction module (i.e., the second prediction module 10).

[0196] Alternatively, each first output input to the second prediction module 10 may be determined qualitatively using prior knowledge regarding the relationship between the first feature computed by the first prediction module 8 and the second feature computed by the second prediction module 10.

[0197] For example, where the second condition is intervertebral disc degeneration, the second prediction module 10 is advantageously configured to receive output from the first prediction module 8 configured to process data of the first condition indicative of the occurrence of Modic-type endplate changes.

[0198] If the artificial intelligence model implemented by the first prediction module 8 is a neural network, the selection of each first output provided to the second prediction module 10 may relate to the selection of weights at a specified depth level of the neural network. In this case, the weights computed by the first layer of the neural network correspond to information of low complexity and are referred to as "low-level" outputs. Furthermore, the weights computed by the last layer of the neural network correspond to information of high complexity and are referred to as "high-level" outputs.

[0199] In another example, the second prediction module 10 is configured to receive as input one or more first feature(s) computed by the first prediction module 8 in addition to the second spine image data set. For example, this configuration is advantageous if the first prediction module 8 is configured to predict the presence of a herniated disc, while the second prediction module 10 is configured to predict the presence of spinal stenosis. In fact, the presence of a herniated disc is considered to be one of the origins of spinal stenosis. Thus, the information that a herniated disc exists will likely make the second prediction module 10 more reliable.

[0200] As another example, the first prediction module 8 is configured to compute a first feature including information on the occurrence of a certain condition as well as an indication regarding this condition. In this case, the second prediction module 10 is configured to receive as input each of the indications computed by the first prediction module 8 in addition to the second spine image data set. For example, this configuration is advantageous if the second prediction module 10 is configured to predict the presence of spinal stenosis and the first prediction module 8 is configured to predict the presence of disc herniation and the location of the disc and spinal canal as corresponding indications. Since such indications are strongly correlated with the condition configured to be predicted by the second prediction module 10, the reliability of such prediction is significantly improved.

[0201] As another example, the first prediction module 8 and the second prediction module 10 are associated with anatomical structures located at different levels along the spine, such as adjacent anatomical structures located along the spine. For example, the first prediction module 8 is configured to predict the presence of spondylolisthesis at a first vertebral level, and the second prediction module 10 is configured to determine the presence of a herniated disc at another vertebral level. This is advantageous because the spine is a mechanically coupled structure of anatomical structures. As a result, biomechanical properties at a given vertebral level (such as spondylolisthesis) will likely have a biomechanical effect on adjacent levels. As a result, the reliability of the second prediction module 10 is improved.

[0202] 2 shows one first prediction module 8 and one second prediction module 10, there may be multiple first prediction modules and / or multiple second prediction modules. In this case, at least one pair including a first prediction module and a second prediction module as described above is defined. Furthermore, the first prediction module may be configured to supply data to multiple second prediction modules.

[0203] Third Prediction Module 12 As shown in figure 2, a double arrow connects the third prediction module 12 with the first prediction module 8. This means that: - Meanwhile, the third prediction module 12 is configured to operate as a second prediction module, thereby computing spine features (different from those computed by the second prediction module 10) based on at least one output of the first prediction module 8 and a third image dataset of the image datasets. On the other hand, module 8 is configured to operate as the second prediction module mentioned above, receiving as input the output of a third prediction module 12 .

[0204] The third prediction module 12 is configured to operate as a second prediction module, but the third image dataset, the output received from the first prediction module, and / or the features computed are different from those corresponding to the second prediction module 10 described above.

[0205] Alternatively, the second prediction module is not provided, in which case module 8 acts as a first prediction module for module 12 and vice versa.

[0206] Fourth Prediction Module 14 2, the fourth prediction module is configured to compute at least one feature related to the target vertebra based on the at least one acquired imaging signal without prior image processing, where the corresponding set of image data is the at least one acquired imaging signal.

[0207] The fourth prediction module 14 may provide an output to and / or receive an output from any of the prediction modules 8, 10, 12. For example, in the embodiment of FIG. 2, the fourth prediction module 14 is configured to provide an output to and receive an output from the second prediction module 10.

[0208] operation During the configuration step, the first prediction module 8 and the second prediction module 10 of the prediction device 6A are configured and each first output to be shared from the first prediction module 8 to the second prediction module 10 is selected.

[0209] If the first prediction module 8 and the second prediction module 10 implement artificial intelligence models, these models are trained, individually or jointly, during the configuration step.

[0210] Next, during an acquiring step, imaging signals representative of at least a portion of the spinal region of interest are acquired.

[0211] The image processing device 4A then performs computations to obtain spinal image data based on the acquired imaging signals.

[0212] Then, during a prediction step, the first prediction module 8 receives the first spine image data set and computes at least one corresponding first output, which comprises a first feature of the examined spine.

[0213] Furthermore, during the prediction step, the second prediction module 10 receives the at least one first output of the first prediction module 8 and the second spine image data set and computes thereon at least one second output, the at least one second output comprising second features of the examined spine.

[0214] For example, the first characteristic and the second characteristic are displayed to a medical practitioner, such as on a display.

[0215] Working Example An exemplary implementation of system 2A is shown in FIG.

[0216] In this case, the imaging signals include sagittal plane sequences T1w and T2w of the lumbar spine of interest.

[0217] These imaging signals are fed to an image processor 4A. Based on these imaging signals, the image processor 4A performs pre-processing, segmentation and post-processing steps to compute regions of interest around the spinal discs, which are provided as outputs. For clarity, only one of the regions of interest around the L4-L5 discs is shown in FIG. 3 according to the two original imaging modalities T1w and T2w. These outputs form the spine image data.

[0218] These regions of interest are provided as input to a first module 8 and a second module 10 configured to perform diagnosis of disc degeneration and endplate degeneration, respectively.

[0219] In the case of disc degeneration, the data of the first condition preferably also includes a grade on the Pfirrmann scale. Furthermore, in the case of endplate degeneration, the data of the first condition preferably also includes a grade on the Modic-type endplate change scale. The same applies to the data of the second condition below.

[0220] Modules 8 and 10 do not have the same input. More precisely, the first spine image data set input to the first module 8 contains only the region of interest according to the sagittal T2w imaging modality. On the other hand, the second spine data set input to the second module 10 contains the region of interest according to the sagittal T2w and T1w imaging modalities. Furthermore, the second spine data set also contains the diagnosis of Pfirrmann disc degeneration performed by the first module 8. This one-way information sharing from module 8 to module 10 was performed because, from a clinical point of view, the grade of the Pfirrmann scale is an explanatory factor for the grade of the Modic scale.

[0221] Second System Embodiment System 2B according to a second system embodiment of the present disclosure is shown in Figure 4. Like system 2A, system 2B is also configured to provide diagnostic assistance for spinal conditions and to implement the methods described above.

[0222] The system 2B comprises an image processing device 4B and a prediction device 6B. The image processing device 4B is configured to perform the receiving step S100 and the first and second processing steps S200 and S400 as described above. The prediction device 6B is configured to perform the first and second prediction steps S300 and S500 as described above.

[0223] Image Processing Unit 4B The second system embodiment differs from the first system embodiment in that the image processing unit 4B comprises at least a first module 20 (also referred to as the "first image processing module") and a second module 22 (also referred to as the "second image processing module"). In the example of Fig. 4, the image processing unit 4B further comprises an additional module 24 (also referred to as the "third image processing module").

[0224] First Image Processing Module 20 The first image processing module 20 is configured to receive a first spine image data set associated with a subject, where the first spine image data set corresponds to a set of acquired imaging signals (e.g., images) representative of a spine of the subject.

[0225] Furthermore, the first image processing module 20 is configured to process at least one first output related to a first anatomical structure of the examined spine based on the spine image data representative of at least a portion of the examined spine, such that the first image processing module 20 is configured to process at least one first output related to a first anatomical structure of the current patient's spine based on a first spine image data set, i.e. a first set of acquired spine imaging signals.

[0226] In this case, the at least one first output includes a first feature of the examined vertebrae.

[0227] For example, in order to computationally process each first output, the first image processing module 20 is configured to implement an artificial intelligence model.

[0228] Furthermore, as described with respect to system 2A, the first output may be any data or parameter (such as weights) that is the result of a calculation performed by the first image processing module 20. Such data may be, for example, a grade (numerical value), a landmark (vector), an area (matrix), a volume (tensor), a segmentation result, a position, a probability, etc. Furthermore, each weight may be a weight of an artificial intelligence model, determined either during or after training of this model.

[0229] Second Image Processing Module 22 The second image processing module 22 is configured to receive a second spine image data set associated with the subject, where the second spine image data set corresponds to a set of acquired imaging signals (e.g., images) representative of the spine of the subject. The second spine image data set may be the same as or different from the first spine image data set.

[0230] The second image processing module 22 is also configured to receive at least one first output of the first image processing module 20. This is indicated in Figure 4 by an arrow pointing from the first image processing module 20 to the second image processing module 22.

[0231] Furthermore, the second image processing module 22 is configured to compute at least one second output related to a second anatomical structure of the examined spine based on the spine image data representative of at least a portion of the examined spine and the at least one first output of the first image processing module 20. Thus, the second image processing module 22 is configured to compute at least one second output related to a second anatomical structure of the current patient's spine based on the at least one first output and the second spine image data set, i.e. the second set of acquired imaging signals. In this case, the at least one second output comprises a second feature of the examined spine.

[0232] For example, to computationally process each second output, the second image processing module 22 is configured to implement an artificial intelligence model.

[0233] Preferably, each of the first and second features represents the shape of a corresponding anatomical structure. More precisely, the first feature representing the shape of the first anatomical structure is related to the second feature representing the shape of the second anatomical structure. In this case, the second anatomical structure is preferably the first anatomical structure or a second anatomical structure of the spine separate from the first anatomical structure, for example an anatomical structure adjacent to the first anatomical structure or an anatomical structure located at a distance from the first anatomical structure.

[0234] For example, for each anatomical structure, the corresponding first feature and / or second feature may include at least one of a corresponding position (or the positions of respective landmarks), a corresponding boundary (i.e., outer limits), a corresponding bounding box(es), associated measurements (such as distances, areas, volumes, and angles within the anatomical structure and between its landmarks), and / or a corresponding label(s).

[0235] Advantageously, the first spine image data set comprises at least one imaging signal acquired according to a first acquisition sequence and the second spine image data set comprises at least one imaging signal acquired according to a second acquisition sequence different from the first acquisition sequence. This allows more information to be taken into account than if only one acquisition sequence were taken into account, thus resulting in more accurate feature computation. This is especially the case when the second anatomical structure is the first anatomical structure.

[0236] Similar to the configuration of the first prediction module and the second prediction module of system 2A, each first output of the first image processing module 20 input to the second image processing module 22 may be data or weights computed by the first image processing module 20.

[0237] Further, as described above with respect to system 2A, each first output of the first image processing module 20 input to the second image processing module 22 may be selected using a qualitative approach or a quantitative approach, or a combination of both approaches.

[0238] As an example, the first image processing module 20 comprises a first stage configured to extract low-level features from the acquired imaging signal and a second stage including a first neural network configured to perform localization and / or identification of the intervertebral disc. Furthermore, the second image processing module 22 comprises a second neural network configured to perform segmentation of the intervertebral disc. In this example, the first and second image processing modules 20, 22 are configured to share information by having co-attention coefficients between their associated layers in the first and second neural networks. These coefficients are learned during training and allow to control how much high-level features from layer i of the first neural network should be shared to layer j of the second module. This is advantageous, especially for segmentation. Indeed, having information about the nature of the anatomical structure (e.g., "the detected vertebra is S1") provides the second neural network with a context as to which vertebra is to be segmented, thereby allowing the second neural network to perform a more accurate segmentation.

[0239] As another example, the first image processing module 20 and the second image processing module 22 are configured to implement a segmentation neural network and a bounding box identification model (also called a "region proposal model"), respectively. In this case, the region proposal model is configured to perform bounding box identification based on the acquired images representing the target vertebrae and the high-level features output by the segmentation model. This feature is advantageous because it speeds up the computational process and improves the quality of the bounding boxes proposed by the region proposal model.

[0240] 4 shows one first image processing module 20 and one second image processing module 22, there may be multiple first image processing modules and / or multiple second image processing modules. In this case, at least one pair including a first image processing module and a second image processing module as described above is defined. Furthermore, the first image processing module may be configured to provide data to multiple second image processing modules.

[0241] Third Image Processing Module 24 As shown in figure 2, a double-headed arrow connects the third image processing module 24 with the first image processing module 20. This means that: On the other hand, the third image processing module 24 is configured to act as the second image processing module 20, thereby computing spine features based on at least one output of the first image processing module 20. On the other hand, module 20 is configured to operate as the above-mentioned second image processing module, receiving as input the output of a third image processing module 24 .

[0242] Alternatively, the second image processing module is not provided, in which case module 20 acts as a first prediction module for module 24 and vice versa.

[0243] Prediction Unit 6B The prediction unit 6B is configured to compute condition data indicative of the occurrence of at least one predetermined condition for the subject's spine based on the first and second features computed by the image processing unit 4B.

[0244] The prediction unit 6B is configured, for example, to implement a known processing pipeline to determine the aforementioned state data.

[0245] operation During a configuration step, the first and second image processing modules 20, 22 of the image processing device 4B are configured and each first output that is shared from the first image processing module 20 to the second image processing module 20 is selected.

[0246] If the first image processing module 20 and the second image processing module 22 implement artificial intelligence models, these models are trained, individually or jointly, during the construction step.

[0247] Next, during an acquiring step, imaging signals representative of at least a portion of the spinal region of interest are acquired, the acquired imaging signals forming a set of spinal image data.

[0248] Then, during an image processing step, the first image processing module 20 receives the first one of the spine image datasets and computes at least one corresponding first output, the at least one first output comprising a first feature of the examined spine.

[0249] Furthermore, during the image processing step, the second image processing module 22 receives the at least one first output of the first image processing module 20 and the second one of the spine image datasets and computes thereon at least one second output, the at least one second output comprising a second feature of the examined spine.

[0250] Next, during a prediction step, the prediction device 6B calculates status data indicative of the occurrence of at least one predetermined condition for the target spine based on at least one first feature and at least one second feature calculated by the image processing device 4B.

[0251] Third System Embodiment System 2C according to a third system embodiment of the present invention is shown in Figure 5. Like systems 2A and 2B, system 2C is also configured to provide diagnostic assistance for spinal conditions and to implement the methods described above.

[0252] The system 2C comprises an image processing device 4C and a prediction device 6C. The image processing device 4C is configured to perform the receiving step S100 and the first and second processing steps S200 and S400 as described above. The prediction device 6C is configured to perform the first and second prediction steps S300 and S500 as described above.

[0253] Image processing device 4C is configured similarly to image processing device 4B of system 2B and includes at least a first module 30 and a second module 32, and preferably includes a third module 34. More precisely, modules 30, 32, and 34 are similar to modules 20, 22, and 24, respectively, of image processing device 4B of system 2B.

[0254] In this case, the spine image data processed by image processor 4C includes acquired imaging signals representative of at least a portion of the subject spine.

[0255] Moreover, prediction device 6C is configured similarly to prediction device 6B of system 2B, including at least a first module 40 and a second module 42, and preferably including a third module 44 and / or a fourth module 46. More precisely, modules 40, 42, 44, and 46 are similar to modules 8, 10, 12, and 14, respectively, of prediction device 6B of system 2B.

[0256] In this case, the spine image data processed by the prediction device 6C is computationally processed by modules 30, 32, and 34 and preferably includes at least first and second features related to the shape of at least one anatomical structure of the target spine.

Claims

1. 1. A computer-implemented method for providing a diagnostic aid for diagnosing a vertebral condition in a spine of a subject by providing at least one characteristic of the spine of the subject, the method comprising: - a receiving step (S100), in which at least first and second imaging signals associated with the object are received, the at least first and / or the second imaging signals being representative of at least a portion of a spine of the object; - a first processing step (S200), in which a first spine data set is computed on the basis of said at least first imaging signals; - a first prediction step (S300), in which at least one first output related to a first anatomical structure of the subject's spine is computed based on the first spine data set, said at least one first output comprising a first feature of the subject's spine; - a second processing step (S400), in which a second spine data set is computed on the basis of said at least second imaging signals; - a second prediction step (S500), in which at least one second output related to a second anatomical structure of the subject spine is computed based on the second spine data set and on the at least one first output, said at least one second output comprising a second feature of the subject spine representative of at least one spinal condition, A method comprising:

2. 2. The computer-implemented method of claim 1, wherein the first processing step (S200) and / or the second processing step (S400) comprises a pre-processing step of computationally processing images from the at least first and / or second imaging signals, preferably the computationally processed images being successive slices of at least a first and / or second portion of the subject's spine representing a 3D view of the at least a first and / or second portion of the subject's spine.

3. 3. The computer-implemented method of claim 2, wherein the pre-processing step also computes at least one of the following on the image: resizing, intensity transformation, artifact correction, bias correction, and / or normalization of pixel intensity distribution.

4. 4. A computer-implemented method according to claim 2 or 3, wherein the pre-processing step is followed by an anatomical structure segmentation step in which the images are processed to identify boundaries of predetermined anatomical structures.

5. 5. The computer-implemented method of claim 4, wherein the anatomical structure segmentation step is performed by a deep learning model trained on a library of reference spine images, the deep learning model outputting a set of raw 2D segmentation masks, each mask associated with a corresponding anatomical structure appearing respectively in the pre-processed images, the deep learning model being preferably a convolutional neural network.

6. After said anatomical structure segmentation step, - a post-processing step on each set of raw 2D segmentation masks to remove false positives and artifacts; and a labelling step, in which the predefined anatomical structures identified in the segmented image are labelled; 6. The computer-implemented method of claim 4 or 5, followed by:

7. 7. The computer-implemented method according to any one of claims 4 to 6, wherein the predetermined anatomical structures include vertebrae and / or intervertebral discs and, for each segmented vertebra in the images acquired after the anatomical segmentation step, a vertebral centroid is computed, preferably the vertebral centroid is computed as the centre of mass of the segmented vertebra based on the corresponding segmentation mask, and, respectively, for each intervertebral disc in the images acquired after the anatomical segmentation step, a disc centroid is computed, preferably the disc centroid is computed as the centre of mass of the segmented disc.

8. 8. The computer-implemented method of any one of claims 1 to 7, wherein the at least one first output is data or parameters of the results of calculations performed on the first spine data set, such as grades (numerical values), landmarks (vectors), areas (matrix), volumes (tensors), segmented results, positions, probabilities, or weights of an artificial intelligence model, such as a neural network, determined during or after its training to perform the calculations.

9. 9. The computer-implemented method of claim 1, wherein the first features include first condition data indicative of an occurrence of a predetermined first condition in the first anatomical structure.

10. 10. The computer-implemented method of any one of claims 1-9, wherein the first condition is one of: Modic type endplate change, Schmorl's nodes, anterior spondylolisthesis, posterior spondylolisthesis, lateral spondylolisthesis, disc degeneration, hypolordosis, hyperlordosis, scoliosis, disc herniation (symmetric bulging, asymmetric bulging protrusion and prolapse) and its location, separation conditions, nerve root compression conditions, spinal stenosis and its origin, lateral recess stenosis and its origin, neural foraminal stenosis and its origin, compression fractures and their acute conditions, paraspinal muscle atrophy, fat involution of paraspinal muscles, facet arthropathy and its origin, tumor, infection, pain, spondylosis.

11. 11. The computer-implemented method of claim 1, wherein the second features include second condition data indicative of an occurrence of a predetermined second condition in the second anatomical structure.

12. 12. A computer-implemented method according to claim 9, wherein the occurrence of the second condition is related to the occurrence of the first condition.

13. The computer-implemented method of claim 12 , wherein the second anatomical structure is the first anatomical structure.

14. 13. The computer-implemented method of claim 12, wherein the second anatomical structure is separate from the first anatomical structure, the second anatomical structure being adjacent to the first anatomical structure or located away from the first anatomical structure.

15. The computer-implemented method of any one of claims 1 to 14, wherein the at least one first output represents a relationship between the first feature and the second feature.

16. 16. A computer-implemented method according to claim 9, wherein the first condition data indicates the occurrence of Modic-type endplate changes and the second condition is disc degeneration.

17. 16. The computer-implemented method of claim 15, wherein the first and second prediction steps (S300, S500) are performed by a first and a second neural network, respectively, the first neural network being trained to generate the at least one first output related to the first anatomical structure of the subject's spine based on the first spine data set, and the second neural network being trained to generate the at least one second output related to the second anatomical structure of the subject's spine based on the second spine data set and the at least one first output, the at least one first output related to a weight at a particular depth level of the first neural network.

18. 18. The computer-implemented method of claim 15 or 17, wherein the first prediction step (S300) is configured to predict the presence of a herniated disc and the second prediction step (S500) is configured to predict the presence of spinal stenosis.

19. 18. The computer-implemented method of claim 15 or 17, wherein the first prediction step (S300) is configured to compute first features comprising information on the occurrence of a given condition as well as hints for this condition, and the second prediction step (S500) is configured to compute the at least one second output based on the second spine data set and the at least one first output comprising the first features and the hints.

20. A system (2A, 2B, 2C) for providing a diagnostic aid for diagnosing a vertebral condition in a spine of a subject by providing at least one characteristic of said spine, said system being configured to perform a method according to any one of claims 1 to 19, - a processing device (4A, 4B, 4C) configured to receive at least first and second imaging signals associated with the object, said at least first and / or second imaging signals representing at least a part of a spine of the object, and configured to compute first and second spine data sets, respectively, based on said at least first and second imaging signals; - a prediction device (6A, 6B, 6C) configured for computing at least one first output related to a first anatomical structure of the subject's spine based on the first spine data set, said at least first output comprising a first feature of the subject's spine, and configured for computing at least one second output related to a second anatomical structure of the subject's spine based on the second spine data set and the at least one first output, said at least second output comprising a second feature of the subject's spine representative of at least one spinal condition; A system (2A, 2B, 2C) comprising:

Images