Method for generating spinal cord synthetic MRI data

Abstract

The present invention concern the field of automated image generation, in particular of the spinal cord. More particularly, it relates to a method for generating an image of a spinal cord of a subject, said image displaying for assessing spinal cord atrophy, a computer program and computer-readable storage medium. Moreover, it also relates to a method for training an image classification algorithm using images with at least one atrophy rate of the spinal cord. The present invention also contemplates, in general, a method for assessing spinal atrophy, a method for assisting in therapy of an atrophy related disease or disorder and a method for monitoring an atrophy rate of a spinal cord. Other fields of application of the present invention, however, are feasible.

Description

[0001] Method for generating spinal cord synthetic MRI data

[0002] Technical Field

[0003] The present invention concern the field of automated image generation, in particular of the spinal cord. More particularly, it relates to a method for generating an image of a spinal cord of a subject, said image displaying spinal cord atrophy, a computer program and computer- readable storage medium. Moreover, it also relates to a method for training an image classification algorithm using images with at least one atrophy rate of the spinal cord. The present invention also contemplates, in general, a method for assessing spinal atrophy, a method for assisting in therapy of an atrophy related disease or disorder and a method for monitoring the atrophy rate of the spinal cord. Other fields of application of the present invention, however, are feasible.

[0004] Background art

[0005] The spinal cord (SC) plays a crucial role in the assessment of various neurological disorders, including multiple sclerosis (MS) [Bischof et al., 2022; Rocca et al., 2022], amyotrophic lateral sclerosis (ALS) [El Mendili et al., 2019; Wendebourg et al., 2024], and SC injury (SCI) [Lundell et al., 2011; Trolle et al., 2023],

[0006] In MS, SC atrophy is particularly prominent in the more debilitating progressive forms of the disease and is considered among the strongest predictor of functional disability [Mina et al., 2021; Tsagkas et al., 2018], In addition, cord atrophy seems to be unrelated to cerebral atrophy suggesting partially independent patterns of neurodegeneration in these two compartments [Cohen et al., 2012; Ruggieri et al., 2015], In ALS, a notable reduction in the cross-sectional area (CSA) of the cervical SC has been reported in patients compared to healthy individuals [Barry et al., 2022], The extent of cervical atrophy has been found to be predictive of shorter life expectancy in ALS patients and found to correlate with functional deficits originating from the SC [de Albuquerque et al., 2017; Grolez et al., 2018], In SCI, there is a rapid and progressive degeneration of the SC, which is further aggravated by a neuro-immune response [Azzarito et al., 2020; Van Broeckhoven et al., 2021],

[0007] With the advancement of MRI scanners, obtaining high-resolution images of the SC, particularly in the cervical region, has become relatively straightforward. By performing semiautomated analysis on these images, either at a specific level (e.g. C1-C2 disk) or along an extended segment of the cord (C2-C5), it is possible to reliably detect atrophy rates of a few percent per year in relatively short-term studies with reasonably sized cohorts of subjects.

[0008] In the last decade, several research groups have developed semi-automated methods for measuring atrophy, utilizing segmentation techniques based on the contrast between the SC tissue and the surrounding cerebrospinal fluid (CSF). The segmentation can be obtained using a wide range of techniques. The most popular are a semi-automatic method based on an active surface (AS) model [Horsfield et al., 2010] and two fully automated methods based on propagation segmentation (PropSeg) [De Leener et al., 2014] and on convolutional neural networks (deepseg) [Gros et al., 2019], respectively. Both PropSeg and deepseg are implemented in the Spinal Cord Toolbox (SCT) [De Leener et al., 2017],

[0009] The approach of extrapolating longitudinal changes from cross-sectional measurements involves numerically subtracting cross-sectional areas obtained separately at two different time points. However, the relatively high measurement noise and low reproducibility associated with segmentation-based methods when measuring the small structure of the SC can affect the accuracy and reliability of the results. Additionally, in a multi-center setting where different imaging protocols and scanners are used, there can be significant variability in the acquired images, making it more challenging to detect small absolute changes in SC area.

[0010] To overcome the obstacles and enhance the assessment of SC atrophy, registration-based techniques have been developed, such as the Generalized Boundary Shift Integral (GBSI) [Prados et al., 2020] and Reg [Valsasina et al., 2022], Unlike segmentation-based techniques, these approaches involve capturing intensity changes in the cord profile over time, resulting in more reliable and consistent measurements of longitudinal changes. By employing registration-based techniques instead of segmentation-based ones, the influence of measurement noise and variability in imaging protocols can be reduced, potentially yielding more reliable and clinically meaningful results. In the context of MS, there is compelling evidence suggesting the presence of preclinical or subclinical disease progression characterized by SC atrophy [Mina et al., 2021], which occurs before the emergence of clinical symptoms [Bischof et al., 2022], It is theorized that the loss of SC volume initiates during the early phase of the disease and precedes the clinical signs of progression [Rocca et al., 2019; Zeydan et al., 2018], For this reason, there is a recognized need to develop reliable imaging analysis techniques that can be automated for use in clinical trials. Such advancements would be valuable in assessing the effectiveness of treatments aimed at slowing down the neurodegenerative mechanisms of progression. By identifying subclinical progression, there is potential to initiate timely interventions that could help prevent or delay the onset of clinical progressive MS [Ontaneda et al., 2023], An unmet need in the field of SC imaging analysis is the availability of ground truth data that accurately simulates longitudinal deformations, specifically atrophy. The lack of such data poses a challenge in evaluating and validating algorithms and techniques designed to assess SC atrophy over time. Having access to ground truth data that faithfully captures the progression of atrophy would enable researchers and developers to benchmark their methods against a reliable reference, ultimately enhancing the accuracy and reliability of SC analysis. Addressing this unmet need is crucial to advancing the field and facilitating the development of more effective diagnostic tools and interventions.

[0011] Different methodologies have been developed for generating artificially induced spinal cord atrophy datasets. The approach described by Prados et al. [Prados et al., 2020], involves an anisotropic axial shrinkage on the cord-straightened image. In another recent study, Bautin and Cohen- Adad conducted a recent study in which T1 -weighted images were rescaled isotropically using different scaling factors. They also simulated patient repositioning by applying random rigid transformations that included rotations and translations [Bautin and Cohen- Adad, 2021],

[0012] These approaches served as valuable resources for studying and evaluating the performance of various segmentation-based algorithms. However, one notable limitation of these assessments is the use of a global or isotropic scaling. This scaling ensures that all anatomical structures are resized while preserving their relative proportions with respect to the SC. However, in a realistic scenario, SC volume decreases, but not the surrounding bones and muscles. In the case of segmentation-based methods, where each follow-up scan is treated as an independent readout, this scaling approach might be considered acceptable. However, for registration-based methods, which heavily rely on the registration between scans, the global scaling effect is typically canceled out in the processing pipeline. Therefore, the use of isotropic scaling in the assessment of registration-based algorithms is not considered acceptable due to its potential to invalidate the results.

[0013] There is hence still a need of providing suitable scaling means for generating spinal cord atrophy datasets.

[0014] Problem to be solved

[0015] The technical problem underlying the present invention may be seen as the provision of means and methods for complying with the aforementioned needs. The technical problem is solved by the embodiments characterized in the claims and herein below.

[0016] It is in particular desirable to establish a comprehensive workflow for generating synthetic data that incorporates an artificial rate of cord atrophy. This synthetic data will be instrumental in validating analysis tools designed for assessing atrophy. Moreover, it is desirable to provide means and methods for accurately quantifying volumetric changes in the upper cervical cord over time. By accomplishing these objectives, a better understanding an improvement of SC analysis techniques can be achieved and a contribution to the development of improved diagnostic tools can be made.

[0017] Summary

[0018] This problem is addressed by a computer-implemented method for generating an image of a spinal cord of a subject for assessing spinal atrophy with the features of the independent claims. Advantageous embodiments which might be realized in an isolated fashion or in any arbitrary combinations are listed in the dependent claims as well as throughout the specification.

[0019] As used in the following, the terms “have”, “comprise” or “include” or any arbitrary grammatical variations thereof are used in a non-exclusive way. Thus, these terms may both refer to a situation in which, besides the feature introduced by these terms, no further features are present in the entity described in this context and to a situation in which one or more further features are present. As an example, the expressions “A has B”, “A comprises B” and “A includes B” may both refer to a situation in which, besides B, no other element is present in A (i.e. a situation in which A solely and exclusively consists of B) and to a situation in which, besides B, one or more further elements are present in entity A, such as element C, elements C and D or even further elements.

[0020] Further, it shall be noted that the terms “at least one”, “one or more” or similar expressions indicating that a feature or element may be present once or more than once typically will be used only once when introducing the respective feature or element. In the following, in most cases, when referring to the respective feature or element, the expressions “at least one” or “one or more” will not be repeated, non-withstanding the fact that the respective feature or element may be present once or more than once.

[0021] Further, as used in the following, the terms "preferably", "more preferably", "particularly", "more particularly", "specifically", "more specifically", “typically”, “more typically” or similar terms are used in conjunction with optional features, without restricting alternative possibilities. Thus, features introduced by these terms are optional features and are not intended to restrict the scope of the claims in any way. The invention may, as the skilled person will recognize, be performed by using alternative features. Similarly, features introduced by "in an embodiment of the invention" or similar expressions are intended to be optional features, without any restriction regarding alternative embodiments of the invention, without any restrictions regarding the scope of the invention and without any restriction regarding the possibility of combining the features introduced in such way with other optional or non-optional features of the invention.

[0022] The present invention contemplates a computer-implemented method for generating an image of a spinal cord of a subject said image displaying spinal atrophy.

[0023] The method comprises the following method steps which, specifically, may be performed in the given order. Still, a different order is also possible. It is further possible to perform two or more of the method steps fully or partially simultaneously. Further, one or more or even all of the method steps may be performed once or may be performed repeatedly, such as repeated once or several times. Further, the method may comprise additional method steps which are not listed.

[0024] The method comprises the following steps: a) providing a reference image of a spinal cord based on pre-recorded MRI data of said subject, wherein image regions of the reference image relating to the spinal cord are pre-identified; b) generating at least one synthetic image with at least one pre-defined atrophy rate of the spinal cord, wherein the generation of said synthetic images comprises the following steps: bl) stripping at least the image regions from the reference image relating to the spinal cord, thereby generating at least one stripped spinal cord region in the reference image and a cord image; b2) filling said stripped spinal cord region with non-zero pixel values relating to cerebrospinal fluid encircling the image regions of the reference image relating to the spinal cord; b3) scaling the cord image by a pre-defined factor relating to a pre-defined atrophy rate; b4) superimposing the scaled cord image onto the filled spinal cord region; c) retrieving said at least one synthetic image with at least one atrophy rate of the spinal cord.

[0025] The computer-implemented method according to the invention may provide advantageous means for generating spinal cord atrophy datasets. These means may contribute to improving the assessment of spinal cord atrophy as an outcome measure in MS clinical trials and for MS diagnostics. Particularly, addressing the unmet need for ground truth data that accurately simulates longitudinal deformations, such as atrophy, is crucial for evaluating and validating spinal cord analysis algorithms. Providing reliable and accurate means for quantifying volumetric changes in the upper cervical cord over time could enable researchers to benchmark and improve their methods, ultimately enhancing the accuracy and reliability of spinal cord atrophy assessment.

[0026] The term "computer implemented method" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a method involving at least one computer and / or at least one computer network. The computer and / or computer network may comprise at least one processor which is con-figured for performing at least one of the method steps of the method according to the present invention. Specifically, each of the method steps is performed by the computer and / or computer network. The method may be performed completely automatically, specifically without user interaction. The term “computer”, thus, may generally refer to a device or to a combination or network of devices having at least one data processing means such as at least one processing unit. The computer, additionally, may comprise one or more further components, such as at least one of a data storage device, an electronic interface or a human-machine interface.

[0027] The term “generating an image of a spinal cord of a subject” as used herein is a broad term and is to be given their ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term may, specifically, refer, without limitation, to the generating of a synthetic image of a spinal cord by means of at least one algorithm using a reference image of a spinal cord based on pre-recorded magnetic resonance imaging (MRI) data of said subject. The term “pre-recorded MRI data” relates to original MRI data comprising MRI images or a set of MRI images that already exist for said subject or that are historical. Said data is typically captured by magnetic resonance imaging as known in the art. Typically the term “pre-recorded MRI data” refers to a plurality of MR image obtained in a MR scan across axial levels of the spinal cord.

[0028] The term “capturing at least one image” may refer, without limitation, to one or more of imaging, image recording, image acquisition, image capturing. The term “capturing at least one image” may comprise capturing a single image and / or a plurality of images such as a sequence of images. For example, the capturing of the image may comprise recording continuously a sequence or set of images such as a set of magnetic resonance (MR) images reflecting a three-dimensional representation of the spinal cord of a subject. The capturing of the at least one image may typically make use of an MRI scanner as known in the art. Automatic MR image acquisition techniques are known, too. The MR image acquisition may be supported by a processor, and the storing of the images may take place in a data storage device. Alternatively or additionally, the captured images may fully or partially be transferred to a cloud system and the storing of the images may fully or partially take place in a data storage device of the cloud system.

[0029] The term “synthetic image” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary image of a physical object, such as a spinal cord of a subject, which comprises computer- based additional information. The term “image” or “original image” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an arbitrary image of a physical object, such as a spinal cord of a subject comprising pixels and or voxels, and more typically information on the axial level of a spinal cord but without said additional computer-generated information. Typically, the “synthetic image” is based on an original image recorded for the spinal cord of a subject, which comprises computer-based additional information. More typically, said additional information comprises pixels or voxels that replace original pixels or voxels in the original image or that are added to the original pixels or voxels representing the original image.

[0030] The term “spinal cord” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to is a long, thin, tubular structure made up of nervous tissue that extends from the medulla oblongata in the brainstem to the lumbar region of the vertebral column (backbone) of vertebrate animals. In line with the present invention, the term in particular refers to the mammalian spinal cord, more particularly to a human spinal cord.

[0031] The term “spinal cord atrophy” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a wasting of spinal cord tissue; in particular, it may refer to loss of spinal cord substance or volume. The wasting of spinal cord tissue or loss of spinal cord volume may in particular result from neurodegenerative diseases or conditions such as MS, ALS, or spinal cord injuries. “Spinal cord atrophy” typically represents an irreversible state of degeneration that may be manifested both radiographically and pathologically, in particular by a loss of spinal cord substance. Typically, said loss of spinal cord substance can be visualized as a shrinkage in spinal cord volume on MR images.

[0032] The term “subject” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a vertebrate animal and, typically, to a mammal. In particular, the subject is a primate and, most typically, a human. A subject in accordance with the present invention may typically have undergone MR imaging, for example for acquiring MR image data of the spinal cord. A subject in accordance with the present invention may typically be a healthy subject, such as an adult subject with a healthy spinal cord. The term “healthy” as used herein refers to a spinal cord that displays no volume shrinkage or loss of substance.

[0033] The term “displaying spinal cord atrophy” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to displaying or showing or otherwise presenting an image of a spinal cord, said spinal cord exhibiting spinal cord wasting, or loss of spinal cord tissue or substance, as defined elsewhere herein. The image may be a recorded image, i.e. an original MR image of a spinal cord, or it may be a synthetic image of a spinal cord, generated according to the method of the invention.

[0034] The term “reference image of a spinal cord” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to an image or a set of images of a spinal cord based on pre-recorded MRI data of a subject, wherein image regions of the reference image relating to the spinal cord are preidentified. More specifically, said reference image may be generated using image segmentation algorithms know in the art such as propagation segmentation, for example PropSeg (De Leener et al., 2024) and convolutional neural networks for example deepseg (Gros et a. 2019). A typical way of generating the reference image of a spinal according to the invention implements both, propagation segmentation and convolutional neural networks. More typically, this involves using the Spinal Cord Toolbox (SCT, De Leener et al., 2017). Even more typically, the generation of the reference image identifies image regions relating to the spinal cord, preferably thereby resulting in pre-identified image regions that may at least be assigned as spinal cord regions and regions that may be assigned as non-spinal cord. Assigning additional regions, such as white matter, grey matter, cerebrospinal fluid and others, is possible. Typically, at least additional image regions relating to the cerebrospinal fluid are identified. A typical reference image according to the invention may be generated using the PAM50 spinal cord template of the SCT toolbox (De Leneer et al., 2018).

[0035] Moreover, in the reference image of the spinal cord, spine curvature may be removed; specifically, during using image segmentation using known algorithm as specified elsewhere herein. Said spine curvature may be re-introduced into the at least one synthetic image based on the pre-recorded MRI data of said subject, typically when the synthetic least one synthetic image with at least one atrophy rate of the spinal cord is retrieved, such as in or prior to step c) of the computer-implemented method according to the invention. Said re-introducing is advantageous as it may contribute to generate synthetic images that align with the subject's anatomical characteristics. This step advantageously may enhance the usability of the synthetic images for further analysis, comparison, and evaluation within the context of the subject's specific spinal cord characteristics and pathology. The term “image region” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to at least one, typically a plurality of, pixel(s) or voxel(s) relating to a region of an image, typically a MR image, more typically an MR image of the spinal cord. More specifically said regions are automatically classified or identified by use of an image classification algorithm as elsewhere described herein. Specifically, the term “image region” as used herein may refer to at least one of the following, spinal cord, cerebrospinal fluid, axial level.

[0036] The computer-implemented method according to the invention comprises step a) of providing a reference image of a spinal cord based on pre-recorded MRI data of said subject, wherein image regions of the reference image relating to the spinal cord are pre-identified.

[0037] The term “providing a reference image of a spinal cord” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to representing, displaying on a screen, storing in a data storage or cloud, or otherwise making available, preferably to a user and / or to a machine, a reference image or a set of reference images of a spinal cord based on pre-recorded MRI data of a subject. In line with the present invention, the reference image is typically displayed on a screen and stored on a data storage device for further processing.

[0038] The term “pre-recorded MRI data” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to existing MRI data of a spinal cord of a subject comprising at least one MR image, typically a set of MR images. In particular said MRI data that has been recorded prior to performing the computer implemented method according to the present invention. In some embodiments, the method may also comprise prior to step a) the step of recording MRI data of a spinal cord of a subject in order to obtain a pre-recorded MRI data of a spinal cord. However, typically the method is an ex vivo method carried out on an existing dataset of MRI data from a subject which does not require any physical interaction with the said subject.

[0039] Moreover, step a) may comprise retrieving said reference image from any of the following: a database, a cloud, a computer, a data storage device; or wherein step a) comprises generating said reference image by an image classification algorithm. The term "retrieving an image" as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. “Retrieving an image” in the context of the application may refer to retrieving a reference image and / or retrieving a synthetic image. The term specifically may refer, without limitation, to obtaining said image in any suitable form, typically from the computer, in particular for making said image available for example to a user or for further processing by a machine, such as a computer, data storage device, screen or printer or the like. The retrieved image has typically been processed by a computer. In line with the invention, a retrieved image is at least one synthetic image displaying at least one atrophy rate of the spinal cord.

[0040] The term “image classification” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of categorizing regions of the image into at least two categories, such as a region relating to the spinal cord or another image region. Typically, the pre-identified regions of the reference image are pre-classified image regions by an image classification algorithm. The classification may be a differentiation and / or segmentation. The term “classification algorithm” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to at least one algorithm designed for performing the classification. The classification algorithm may use at least one parametric model or at least one non-parametric model. The classification algorithm may take the image and / or regions extracted from the image as input. For example, the extracted regions may be shape and / or color statistics. The classification algorithm may use at least one classification algorithm selected from the group consisting of at least one convolutional neural network (CNN) technology, deconvolutional neural network (DNN) technology, decision trees, point vector, at least one transformer based model, random forest, K-Nearest-Neighbor, and the like.

[0041] The classification algorithm may comprise at least one trained machine learning model. The term “machine learning model” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a mathematical model which is trainable on at least one training dataset using machine learning, in particular deep learning or other form of artificial intelligence. The term “machine leaming” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a method of using artificial intelligence (Al) for automatically model building. The method may comprise at least one training step. The training step may comprise training the machine learning model based on the training dataset. The term “training” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to a process of building the trained machine learning model, in particular determining parameters, in particular weights, of the model. The training may comprise determining and / or updating parameters of the model. The trained machine learning model may be at least partially data driven. As used herein, the term “at least partially data-driven” is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to the fact that the model comprises data-driven model parts. The model may comprise additional other model parts such as based on physico-chemical laws. The training may be performed on historical MR images. The training dataset may be generated by manual classification of the image regions of the historical MR images, e.g. into at least two categories.

[0042] Typically, the image classification algorithm suitable for use in the present invention comprises or performs at least the following steps: identifying the spinal cord region in said pre-recorded MRI data of said subject, more typically, by partitioning the MR images into different segments comprising sets of voxels assigned to the spinal cord; registering the identified spinal cord region to a template, eve more typically to a straightened template, for example PAM50 template, thereby generating the reference image; and / or assigning axial levels to the spinal cord region.

[0043] More typically, the image classification algorithm uses at least one classification algorithm selected from the group consisting of: at least one convolutional neural network (CNN) technology, deconvolutional neural network (DNN) technology, decision trees, point vector, random forest, K-Nearest-Neighbor.

[0044] Additional steps may for example comprise: vertebral labeling for generating the reference image, which may typically be non-linear and are given in the examples in more detail. Moreover, based on common spinal cord image tools, such as SCT, a non-linear mapping may be applied in the generation of the reference image of a spinal cord. This mapping function may be inverted later on to align the synthetic images based on the reference image onto the pre-recorded MRI data.

[0045] Step b) of generating at least one synthetic image with at least one pre-defined atrophy rate of the spinal cord comprises the following steps: bl) stripping at least the image regions from the reference image relating to the spinal cord, thereby generating at least one stripped spinal cord region in the reference image and a cord image; b2) filling said stripped spinal cord region with non-zero pixel values relating to cerebrospinal fluid encircling the image regions of the reference image relating to the spinal cord; b3) scaling the cord image by a pre-defined factor relating to a pre-defined atrophy rate; and b4) superimposing the scaled cord image onto the filled spinal cord region.

[0046] Said generating at least one synthetic image is typically an automated process step that does not involve the interaction with a user. Said synthetic image generation more typically involves a sequence of image manipulation steps applied to the reference image of the spinal cord. Even more typically, the generation of the synthetic image starts with step bl) stripping at least the image regions from the reference image relating to the spinal cord, thereby generating at least one stripped spinal cord region in the reference image and a cord image. Said step may involve removing the image regions pre-identified as relating to the spinal cord resulting in said at least one stripped spinal cord region in the reference image. Said stripped spinal cord region may by identified as relating to the cerebrospinal fluid.

[0047] Particularly, said stripping in step bl) may comprise stripping the image regions relating to the spinal cord from the reference image, or step bl) may comprise identifying further image regions relating to a dilated spinal cord, and stripping the image regions relating to the spinal cord and the further image regions relating to the dilated spinal cord from the reference image.

[0048] Specifically, said identifying further image regions relating to a dilated spinal cord may comprise the dilation of the regions pre-identified as relating to the spinal cord, for example by applying a circular structuring element with a suitable diameter, such as four pixels, to achieve the dilation effect. This may be achieved by using the function imdilate in MATLAB or similar algorithms. Hence, the further image regions relating to the dilated spinal cord are typically identified by applying a circular structuring element with a diameter of a specified number of pixels, such as four pixels, for removing any non-cord voxels, more typically thereby identifying a dilated spinal cord region, typically on each axial level.

[0049] Moreover, in step bl) a cord image is generated. Said cord image typically comprises, more typically consists of, the image regions pre-identified as relating to the spinal cord.

[0050] As a further step, said generating at least one synthetic image comprises step b2) of filling said stripped spinal cord region with non-zero pixel values relating to cerebrospinal fluid encircling the image regions of the reference image relating to the spinal cord. Typically, said stripped spinal cord region corresponds to the spinal canal. The filling may be achieved by selecting voxels of the cerebrospinal fluid encircling the image regions of the reference image relating to the spinal cord and interpolating the inward. More typically, filling said stripped spinal cord region with non-zero pixel values comprises using at least one mathematical algorithm, wherein the mathematical algorithm comprises using at least one filling function. Still more typically, said filling said stripped spinal cord region with non-zero pixel values comprises at least one mathematical algorithm based on computing a discrete Laplacian over the stripped spinal cord region, typically followed by calculating the Dirichlet distribution, to fill said region. Said filling may specifically be achieved by using the function regionfill in MATLAB or similar algorithms.

[0051] Even further, said generating at least one synthetic image comprises step b3) of scaling the cord image by a pre-defined factor relating to a pre-defined atrophy rate. Typically, said scaling of the cord image is isotropic scaling. More typically, the scaling factor is in the range of 0.9975 to 0.95, still more typically relating to an atrophy rate in the range of 0.5% to 10 % compared to the cord image. The scaling of the cord image in particular results in the generation of a scaled cord image. Specifically, said scaling refers to downscaling the cord image, in particular relating to a pre-defined atrophy rate. The pre-defined scaling factor more specifically downscales the cord image. Said downscaling of the cord image still more specifically refers to an isotropic downscaling, in particular in a desired range, for example to 0.9975 to 0.95 in comparison to an original image size of 1.0. Typically, said downscaling relates to an atrophy rate in the range of 0.5 % to 10 % compared to the cord image, More typically, said downscaling may be performed using a pre-defined scaling factor between 0.9975 and 0.95.

[0052] The generating at least one synthetic image according to the invention further comprises step b4) of superimposing the scaled cord image onto the filled spinal cord region, in particular the filled spinal canal. Typically, this process comprises replacing image regions assigned as cerebrospinal fluid, specifically in the filling step, by image regions of the scaled cord image assigned as spinal cord. More typically, the said superimposing step comprises replacing voxels that assigned to the cerebrospinal fluid by voxels of the scaled cord image assigned as cord voxels. The scaled cord image typically refers to a downscaled cord image.

[0053] In particular, the computer-implemented method for generating an image of a spinal cord of a subject, said image displaying spinal cord atrophy may further comprise step b5) of repeating steps b3) and b4) thereby generating a set or a plurality of synthetic images of the spinal cord.

[0054] More particularly, the computer-implemented method for generating an image of a spinal cord of a subject, said image displaying spinal cord atrophy may further comprise step b5) of repeating steps b3) and b4) for different atrophy rates thereby generating a plurality of synthetic images with different atrophy rates of the spinal cord.

[0055] Further the computer-implemented method according to the invention comprises step c) retrieving said at least one synthetic image with at least one atrophy rate of the spinal cord. Said retrieving may refer to obtaining said synthetic image in any suitable form, typically from the computer, in particular for making said image available for example to a user or for further processing by a machine, such as a computer, data storage device, screen or printer or the like.

[0056] Moreover, the computer-implemented method for generating an image of a spinal cord of a subject, may comprise a step of introducing noise, typically by introducing noise based on a Rician distribution and / or on intensity inhomogenities. Such routines are known in the art and for example disclose in Ganzetti et al., Front. Neuroinform., 2016. 10: 10.

[0057] The computer-implemented method for generating an image of a spinal cord of a subject may further comprise a step of monitoring atrophy rates in a subject based on said synthetic image or synthetic images.

[0058] Still further, the computer-implemented method according to the invention may comprise using said synthetic image or images in one of the following applications:

[0059] - diagnosis of spinal atrophy;

[0060] - prognosis of spinal atrophy;

[0061] - monitoring spinal atrophy; - training of an image classification algorithm for identifying spinal atrophy;

[0062] - quantifying the level of atrophy in a clinical setting;

[0063] - image classification; and / or

[0064] - image segmentation.

[0065] Even further steps for generating at least one synthetic image with at least one pre-defined atrophy rate of the spinal cord are possible, such as at least one of the following steps: displaying said retrieved synthetic image on a screen; evaluating said retrieved synthetic image using at least one evaluation operation; and / or transferring said retrieved synthetic image to at least one data storage device.

[0066] The term “assessing spinal cord atrophy based on said synthetic images” as used herein is a broad term and is to be given its ordinary and customary meaning to a person of ordinary skill in the art and is not to be limited to a special or customized meaning. The term specifically may refer, without limitation, to assessing whether an image of a spinal cord of a subject is exhibiting spinal cord atrophy, or not. Accordingly, assessing as used herein includes identifying spinal cord atrophy in an image, identifying improvement of the said cognition and movement disease or disorder or one or more symptoms accompanying it, monitoring the said cognition and movement disease or disorder or one or more symptoms accompanying it, determining efficacy of a therapy of the said cognition and movement disease or disorder or one or more symptoms accompanying it, and / or diagnosing the said cognition and movement disease or disorder or one or more symptoms accompanying it. As will be understood by those skilled in the art, such an assessment, although preferred to be, may usually not be correct for 100% of the investigated subjects. The term, however, requires that a statistically significant portion of subjects can be correctly assessed and, thus, identified as suffering from the cognition and movement disease or disorder. Whether a portion is statistically significant can be determined without further ado by the person skilled in the art using various well known statistic evaluation tools, e.g., determination of confidence intervals, p- value determination, Student's t-test, Mann- Whitney test, etc.. Details may be found in Dowdy and Wearden, Statistics for Research, John Wiley & Sons, New York 1983. Typically envisaged confidence intervals are at least 50%, at least 60%, at least 70%, at least 80%, at least 90%, at least 95%. The p-values are, typically, 0.2, 0.1, 0.05. Thus, the method of the present invention, typically, aids the assessment of cognition and movement diseases or disorders by providing a means for evaluating a dataset of activity measurements. The present invention moreover contemplates a method for training an image classification algorithm using images with at least one atrophy rate of the spinal cord comprising generating one or more images of a spinal cord of a subject by the steps as defined in the computer implemented method for generating an image of a spinal cord of a subject, said one or more images displaying spinal cord atrophy. The method typically comprises training said image classification algorithm using said one or more images of a spinal cord of a subject displaying spinal cord atrophy as a training dataset. The training may more typically comprise using a plurality of said images as a training dataset. The one or more images displaying spinal cord atrophy may in particular be obtained by repeating steps b) to c) of the computer-implemented method for generating an image of a spinal cord of a subject for different predefined atrophy rates; i.e. for more than one pre-defined atrophy rate. More particular, said one or more images displaying spinal cord atrophy display different rates of pre-defined spinal cord atrophy.

[0067] Further, the present invention relates to a dataset for training of an image classification algorithm, specifically an image classification algorithm configured for identifying spinal atrophy, obtained or obtainable by the computer-implemented method according to the invention as defined elsewhere herein. The term "machine learning model" refers to a trainable computer-implemented architecture, particularly a trainable statistical model, that applies artificial intelligence to automatically determine a representative result, particularly the confidence value. As generally used, the term “training” refers to a process of determining adjustable parameters of the machine learning model using a training data for generating a trained machine learning model. The training may comprise at least one optimization or tuning process, wherein a best parameter combination is determined. The training is carried out to improve the capability of the machine learning model to determine the representative result, particularly the confidence value, by analyzing at least a portion of the training data. The confidence value, typically referred to as confidence interval, may commonly be determined by the person of skill in the art using well known statistic evaluation tools, e.g., determination of confidence intervals, p-value determination, Student's t-test, Mann- Whitney test, etc.. Details may be found in Dowdy and Wearden, Statistics for Research, John Wiley & Sons, New York 1983. Typically envisaged confidence intervals are at least 50%, at least 60%, at least 70%, at least 80%, at least 90%, at least 95%.

[0068] As generally used, the term “training dataset” refers to data that is used for training of the machine learning model. The parameters of the machine learning model may, thus, be adjusted, when at least a portion of the training dataset is analyzed by the machine learning model. The training data, in that sense, may comprise at least one “known” information and the machine learning model may, further, be configured for a training of determining information that corresponds to the known information as accurate as possible, or, in other words, the determined information may deviate from the known information as little as possible. The training dataset may comprise real data and / or synthetic data, in particular original and / or synthetic MRI data. Typically, said training dataset comprises or consists of a plurality of synthetic MR images displaying spinal cord atrophy, more typically the training dataset has a size in the order of thousands of images, even more typically said training dataset comprises or consists of at least 1000, still even more typically 10’000 of synthetic MR images displaying spinal cord atrophy.

[0069] Moreover, the present invention relates to a method for assessing spinal atrophy comprising generating an image of a spinal cord of a subject displaying spinal cord atrophy by performing the method for generating an image of a spinal cord of a subject as defined elsewhere herein, further comprising comparing said synthetic image or plurality of synthetic images of the spinal cord to a reference, typically to an earlier set of synthetic images or to a reference MRI dataset; and more typically determining the atrophy rate based on said comparison.

[0070] Still further, a method for assisting in therapy of an atrophy related disease or disorder comprising generating at least one image displaying spinal cord atrophy by performing the method for generating an image of a spinal cord of a subject as defined elsewhere herein, further comprising comparing said synthetic image of the spinal cord to a reference, typically to an earlier synthetic image or set of synthetic images or to a reference MRI dataset; and more typically determining a change in the atrophy rate based on said comparison is contemplated by the present invention.

[0071] Furthermore, the invention relates to a method for monitoring an atrophy rate of a spinal cord, typically in a subject suspected or known to suffer from an atrophy causing disease or disorder, comprising generating a set of synthetic images of a spinal cord of a subject, said images displaying spinal cord atrophy, wherein the images are generated performing the method for generating an image of a spinal cord of a subject as defined elsewhere herein, further comprising comparing said synthetic image of the spinal cord to a reference, typically to an earlier synthetic image or set of synthetic images or to a reference MRI dataset; and more typically determining a change in the atrophy rate based on said comparison. Also contemplated by the invention is a computer program comprising instructions which, is executed by a computer or a computer network, cause the computer or computer network to perform the method according to the computer-implemented method according to the invention.

[0072] Still the present invention relates to a computer-readable storage medium comprising instructions which, when the instructions are executed by a computer or a computer network, cause the computer or computer network to perform the method according to the computer-implemented method according to the invention.

[0073] Moreover, a non-transient computer-readable medium including instructions that, when executed by one or more processors, cause the one or more processors to perform the computer-implemented method according to the invention is contemplated.

[0074] Further disclosed and proposed herein is a computer program product with program code means stored on a machine-readable carrier, in order to perform the method according to one or more of the embodiments disclosed herein, when the program is executed on a computer or computer network. As used herein, a computer program product refers to the program as a tradable product. The product may generally exist in an arbitrary format, such as in a paper format, or on a computer-readable data carrier and / or on a computer-readable storage medium. Specifically, the computer program product may be distributed over a data network.

[0075] Finally, disclosed and proposed herein is a modulated data signal which contains instructions readable by a computer system or computer network, for performing the method according to one or more of the embodiments disclosed herein.

[0076] Referring to the computer-implemented aspects of the invention, one or more of the method steps or even all of the method steps of the method according to one or more of the embodiments disclosed herein may be performed by using a computer or computer network. Thus, generally, any of the method steps including provision and / or manipulation of data may be performed by using a computer or computer network. Generally, these method steps may include any of the method steps, typically except for method steps requiring manual work, such as providing the samples and / or certain aspects of performing the actual measurements.

[0077] Specifically, further disclosed herein are: - a computer or computer network comprising at least one processor, wherein the processor is adapted to perform the method according to one of the embodiments described in this description,

[0078] - a computer loadable data structure that is adapted to perform the method according to one of the embodiments described in this description while the data structure is being executed on a computer,

[0079] - a computer program, wherein the computer program is adapted to perform the method according to one of the embodiments described in this description while the program is being executed on a computer,

[0080] - a computer program comprising program means for performing the method according to one of the embodiments described in this description while the computer program is being executed on a computer or on a computer network,

[0081] - a computer program comprising program means according to the preceding embodiment, wherein the program means are stored on a storage medium readable to a computer,

[0082] - a storage medium, wherein a data structure is stored on the storage medium and wherein the data structure is adapted to perform the method according to one of the embodiments described in this description after having been loaded into a main and / or working storage of a computer or of a computer network, and

[0083] - a computer program product having program code means, wherein the program code means can be stored or are stored on a storage medium, for performing the method according to one of the embodiments described in this description, if the program code means are executed on a computer or on a computer network.

[0084] Summarizing and without excluding further possible embodiments, the following embodiments may be envisaged:

[0085] 1. A computer-implemented method for generating an image of a spinal cord of a subject, said image displaying spinal cord atrophy, comprising: a) providing a reference image of a spinal cord based on pre-recorded MRI data of said subject, wherein image regions of the reference image relating to the spinal cord are pre-identi- fied; b) generating at least one synthetic image with at least one pre-defined atrophy rate of the spinal cord, wherein the generation of said synthetic images comprises the following steps: bl) stripping at least the image regions from the reference image relating to the spinal cord, thereby generating at least one stripped spinal cord region in the reference image and a cord image; b2) filling said stripped spinal cord region with non-zero pixel values relating to cerebrospinal fluid encircling the image regions of the reference image relating to the spinal cord; b3) scaling the cord image by a pre-defined factor relating to a pre-defined atrophy rate; b4) superimposing the scaled cord image onto the filled spinal cord region; c) retrieving said at least one synthetic image with at least one atrophy rate of the spinal cord.

[0086] 2. The method according to embodiment 1, wherein spine curvature in the reference image of the spinal cord is removed.

[0087] 3. The method according to embodiment 2, wherein the spine curvature is re-introduced into the at least one synthetic image based on the pre-recorded MRI data of said subject.

[0088] 4. The method according to any of the preceding embodiments, wherein step a) comprises retrieving said reference image from any of the following: a database, a cloud, a computer, a data storage device; or wherein step a) comprises generating said reference image by an image classification algorithm.

[0089] 5. The method according to the preceding embodiment, wherein said image classification algorithm comprises at least the following steps:

[0090] - identifying the spinal cord region in said pre-recorded MRI data of said subject, typically by partitioning the MR images into different segments comprising sets of voxels assigned to the spinal cord;

[0091] - registering the identified spinal cord region to a template, typically to a straightened template, for example PAM50 template, thereby generating the reference image;

[0092] - assigning axial levels to the spinal cord region.

[0093] 6. The method according to any of the two preceding embodiments, wherein the image classification algorithm uses at least one classification algorithm selected from the group consisting of: at least one convolutional neural network (CNN) technology, deconvolutional neural network (DNN) technology, decision trees, point vector, random forest, K-Nearest- Neighbor.

[0094] 7. The method according to any of the preceding embodiments, wherein said stripping in step bl) comprises stripping the image regions relating to the spinal cord from the reference image, or wherein step bl) comprises identifying further image regions relating to a dilated spinal cord, and stripping the image regions relating to the spinal cord and the further image regions relating to the dilated spinal cord from the reference image.

[0095] 8. The method according to any of the preceding embodiments, wherein the scaling of the cord image is isotropic.

[0096] 9. The method according to any of the preceding embodiments, wherein the scaling factor is in the range of 0.9975 to 0.95, typically relating to an atrophy rate in the range of 0.5% to 10 % compared to the cord image.

[0097] 10. The method according to any of the preceding embodiments, further comprising at least one of the following steps:

[0098] - displaying said retrieved synthetic image on a screen;

[0099] - evaluating said retrieved synthetic image using at least one evaluation operation; and / or

[0100] - transferring said retrieved synthetic image to at least one data storage device.

[0101] 11. The method according to any of the four preceding embodiments, wherein the further image regions relating to the dilated spinal cord are identified by applying a circular structuring element with a diameter of a specified number of pixels, such as 4 pixels, for removing any non-cord voxels, typically thereby identifying a dilated spinal cord region, typically on each axial level.

[0102] 12. The method according to any of the preceding embodiments, wherein filling said stripped spinal cord region with non-zero pixel values comprises using at least one mathematical algorithm, wherein the mathematical algorithm comprises one or more of: using at least one filling function.

[0103] 13. The method according to the preceding embodiment, wherein filling said stripped spinal cord region with non-zero pixel values comprises at least one mathematical algorithm based on computing a discrete Laplacian over the stripped spinal cord region, typically followed by calculating the Dirichlet distribution, to fill said region.

[0104] 14. The method according to any of the preceding embodiments, further comprising step b5) of repeating steps b3) and b4) for different atrophy rates thereby generating a plurality of synthetic images with different atrophy rates of the spinal cord. 15. The method according to any of the preceding embodiments, further comprising a step of introducing noise, typically by introducing noise based on a Rician distribution and / or on intensity inhomogeneities.

[0105] 16. The method according to any of the preceding embodiments, further comprising a step of assessing spinal atrophy based on said synthetic images.

[0106] 17. The method according to any of the preceding embodiments, further comprising a step of monitoring atrophy rates in a subject based on said synthetic images.

[0107] 18. The method according to any of the preceding embodiments, further comprising using said synthetic images in one of the following applications:

[0108] - diagnosis of spinal atrophy;

[0109] - prognosis of spinal atrophy;

[0110] - monitoring spinal atrophy;

[0111] - training of an image classification algorithm for identifying spinal atrophy;

[0112] - quantifying the level of atrophy in a clinical setting;

[0113] - image classification; and / or

[0114] - image segmentation.

[0115] 19. A dataset for training of an image classification algorithm, specifically an image classification algorithm configured for identifying spinal atrophy, obtained by the method as defined in any of embodiments 1 to 18.

[0116] 20. A method for training an image classification algorithm using images with at least one atrophy rate of the spinal cord comprising generating one or more images of a spinal cord of a subject as defined in the method of any of the preceding embodiments 1 to 19, said one or more images displaying spinal cord atrophy; training the image classification algorithm using said one or more image, or using a plurality of said images, as a training dataset.

[0117] 21. A method for assessing spinal atrophy comprising generating an image of a spinal cord of a subject displaying spinal cord atrophy by performing the method as defined in any of the preceding embodiments 1 to 19, further comprising comparing said synthetic image or plurality of synthetic images of the spinal cord to a reference, typically to an earlier set of synthetic images or to a reference MRI dataset; and more typically determining the atrophy rate based on said comparison. 22. A method for assisting in therapy of an atrophy related disease or disorder comprising generating an image of a spinal cord of a subject, said spinal cord displaying spinal cord atrophy, by performing the method as defined in any of the preceding embodiments 1 to 19, further comprising comparing said synthetic images of the spinal cord to a reference, typically to an earlier set of synthetic images or to a reference MRI dataset; and more typically determining a change in the atrophy rate based on said comparison.

[0118] 23. A method for monitoring an atrophy rate of a spinal cord, typically in a subject suspected or known to suffer from an atrophy causing disease or disorder, comprising generating a set of synthetic images of a spinal cord of a subject, said images displaying spinal cord atrophy, wherein the images are generated by performing the method as defined in any of the preceding embodiments 1 to 19, further comprising comparing said synthetic images of the spinal cord to a reference, typically to an earlier set of synthetic images or to a reference MRI dataset; and more typically determining a change in the atrophy rate based on said comparison.

[0119] 24. A computer program comprising instructions which, is executed by a computer or a computer network, cause the computer or computer network to perform the method according to any one of the preceding embodiments referring to a computer-implemented method.

[0120] 25. A computer-readable storage medium comprising instructions which, when the instructions are executed by a computer or a computer network, cause the computer or computer network to perform the method according to any one of the preceding embodiments referring to a computer-implemented method.

[0121] 26. A non-transient computer-readable medium including instructions that, when executed by one or more processors, cause the one or more processors to perform method according to any one of the preceding embodiments referring to a computer-implemented method.

[0122] All references cited in this specification are herewith incorporated by reference with respect to their entire disclosure content and the disclosure content specifically mentioned in this specification.

[0123] Figure Legends

[0124] Figure 1 illustrates the workflow of the method according to the invention. Figure 2 shows the output of the workflow.

[0125] Figure 3 depicts the characterization of the intensity distribution for different levels of simulated atrophy. Intensity values for different atrophy levels are plotted. Values are extracted within the PAM50 SC mask (SC levels 1-8).

[0126] Figure 4 shows the impact of non-rigid registration from template to subject space. Atrophy defined in template space is preserved in subject space. Intensity value differences are extracted within the PAM50 SC mask (SC levels 1-8)

[0127] Figure 5 depicts images with different noise levels. The noise level is defined as percentage of median SC intensity (extracted on PAM50 template). A noise level 0%, B noise level 1%, C noise level 2%, D noise level 4% , E noise level 8%

[0128] Figure 6 shows the performance of segmentation-based vs registration-based methods.

[0129] Examples

[0130] The following Examples shall merely illustrate the invention. They shall not be construed, whatsoever, to limit the scope of the invention.

[0131] Example 1: Methods

[0132] In this section, we outline the workflow for generating synthetic MR images of the SC. Furthermore, we demonstrate the practical application of this artificial data by highlighting its potential in evaluating the performance of tools designed for analyzing SC atrophy.

[0133] Description of the workflow

[0134] The generation of synthetic MR data follows a two-step workflow that ensures the production of highly precise and accurate synthetic MR images. In this context, we will focus on the Tl-weighted (Tl-w) modality, which is commonly used in clinical and research settings to evaluate tissue structures and to determine important cord geometry measures, such as the cross-sectional area (CSA) [Cohen- Adad et al., 2021a],

[0135] Generation of synthetic MR images in template space

[0136] The initial phase involves a sequence of image manipulation steps applied to the PAM50 spinal cord template. This template establishes a standardized reference image that is commonly utilized to facilitate population-based analyses. This multimodal MRI template of the SC and the brainstem is anatomically compatible with the ICBM152 brain template and uses the same coordinate system [De Leener et al., 2018], The procedure begins with the dilation of the cord's binary mask, a component of the PAM50 template, on each axial level. This is performed using the imdilate function in MATLAB, which takes advantage of a circular structuring element with a diameter of 4 pixels to achieve the dilation effect. Subsequently, any non-cord voxels from the original PAM50 template are removed using the enlarged cord mask from the previous step. Once the cord has been excised from the image, a region-filling method is implemented. For this task, MATLAB's regionfill function is employed. This function selects voxels of cerebrospinal fluid that encircle the excised area and smoothly interpolates them inward, taking cues from the values at the region's outer boundary. The function executes this by computing the discrete Laplacian over the specified area and solving the Dirichlet problem to seamlessly fill in the space. Following the region-filling step, the extracted spinal cord image undergoes isotropic scaling to the desired size, and then it is resliced using trilinear interpolation to reduce artifacts at the junctions between WM and CSF. To finalize the process, the rescaled cord image is carefully superimposed onto the image with the filled canal.

[0137] Upon creating the simulated image that exhibits the specified level of atrophy, there exists the option to introduce noise that adheres to a Rician distribution into the pristine artificial image. Rician noise is the type of thermal noise encountered in MRI scans, which originates from the thermal movement of electrons [Gudbjartsson and Patz, 1995], For the purpose of this simulation, images have been produced with escalating degrees of noise (percentage relative to the median intensity of the SC) according to the following formula:

[0138] NI= (n2+(n+I)2)

[0139] With NI and I as the image with added noise and the noise-free image, respectively, n is a tridimensional array of random numbers drawn from the normal distribution with mean equal to zero and standard deviation: o_l=n_l-median(SC). n_l is referring to the specific noise level (e.g. 1%, 2%, 4%, and 8%).

[0140] Registration of synthetic images from template space to subject space

[0141] After the generation of template-based synthetic images, the second part of the workflow involves a series of image segmentation and registration steps utilizing SCT v6.0 [De Leener et al., 2017], These processing steps are crucial in aligning a real Tl-w image with the corresponding PAM50 template image, from which synthetic images characterized by known rates of artificial atrophy have been constructed. By employing an initial image segmentation step, the SC region of interest is accurately delineated, enabling precise registration with the template image. The registration steps within the SCT ensure optimal alignment between the real and synthetic images. This first step utilize the deep learning-based segmentation algorithm available in SCT [Gros et al., 2019], The sequential steps of the algorithm are as follows: cord detection using a convolutional neural network, centerline detection, patch extraction to exclude regions that are certainly not containing the spinal cord and a final cord segmentation according to a second convolutional neural network applied to the extracted patch.

[0142] Following the segmentation, vertebral labeling is performed [Ullmann et al., 2014], First the SC is straightened by finding, for each point along the cord, the mathematical transformation to go from a curved to a straight centerline. The algorithm computes the orthogonal plane at each point along the centerline, and then creates a straight space in the output using thin- plate spline interpolation. After straightening the cord, the subsequent step in the workflow involves a transformation to align the vertebral levels of the subject with that of the template image. This transformation ensures accurate correspondence between the subject and template images in terms of vertebral anatomy. To achieve this alignment, two labels indicating the first and last available vertebral levels on the real subject image need to be provided. These labels serve as reference points for the transformation, enabling precise matching of vertebral levels between the subject and template.

[0143] After vertebral labeling, the workflow proceeds with a shape-matching transformation [De Leener et al., 2018], This step involves estimating a multi-step non-rigid deformation to align the shape of the subject's cord with the template. By default, the transformation consists of two steps. The first step focuses on handling large deformations, ensuring an initial alignment between the subject's cord shape and the template. This accounts for major structural differences between the subject and template. The second step involves fine adjustments, refining the alignment to achieve a precise matching. This multi-step non-rigid deformation approach enhances the accuracy of the alignment.

[0144] Once the transformations are estimated, we can apply the resulting warping field to the synthetic images in the template space to bring them into to the subject’s native space, thus generating synthetic images that align with the subject's anatomical characteristics. This step enhances the usability of the synthetic images for further analysis, comparison, and evaluation within the context of the subject's specific SC characteristics and pathology.

[0145] Example 2: Method validation In order to evaluate the accuracy of the workflow, a total of six subjects were randomly chosen from the open-access quantitative MRI dataset [Cohen- Adad et al., 2021a], This dataset consists of both single-participant and multi-participant data collected from various centers. The participants underwent scanning following a predefined protocol [Cohen- Adad et al., 2021b], The data was collected, organized, and analyzed using a well-documented procedure, which can be found at https: / / spine-generic.rtfd.io. Out of all the sequences available in the dataset, the defaced 3D sagittal Tlw image was utilized for the purpose of validation. Table 1 provides detailed information regarding the demographics and scanner pa- rameters for the selected subjects.

[0146] Table 1 - Subjects demographics

[0147] Results

[0148] To evaluate the effectiveness of the entire workflow, the intensity distribution within the spinal cord binary mask was examined at different levels of simulated atrophy, as depicted in Figure 3. The first row of Figure 3 demonstrates the relationship between the intensities of the original spinal cord (without atrophy, 100%) and the corresponding intensities of the spinal cord with different levels of induced artificial atrophy (2, 4, 6, 10, and 20%) in the PAM50 template. The second row presents the same information for a representative subject. In both cases, the varying degrees of induced atrophy can be observed. The deviation of intensity from the diagonal line is proportional to the degree of induced atrophy.

[0149] Additionally, to evaluate the impact of the non-rigid registration used to transform the artificial image from PAM50 template space to the original subject space (as described in the workflow in Figure 1), the intensity differences between the spinal cord without atrophy and the cord with a specific atrophy rate were computed. This assessment was conducted for both the PAM50 template and the non-linearly registered PAM50 template in subject space, as shown in Figure 4. Plotting both distributions against each other allows for a better evaluation of the correspondence between the two.

[0150] In order to increase the complexity and generate synthetic images that closely resemble real MR images, Rician noise with varying magnitudes was added to the synthetic image, as demonstrated in Figure 5.

[0151] Figure 6 presents an assessment of the performance of two methods used for quantifying spinal cord atrophy on synthetic images with different levels of atrophy (0, 2, 4, 6, 10, and 20%). The performance of segmentation-based (PC CSA) and registration-based (PC REG) methods is reported for two cervical segments: C1-C2 and C2-C5. Both methods exhibit similar performance with high accuracy.

[0152] In conclusion, the establishment of a comprehensive workflow for generating synthetic data with an artificial rate of cord atrophy will contribute to advancing the field.

[0153] Moreover, continued advancements in imaging techniques and analysis methods, along with the exploration of alternative measurement approaches, will contribute to improving the assessment of spinal cord atrophy as an outcome measure in MS clinical trials.

[0154] Literature cited: de Albuquerque M, et al. (2017): Longitudinal evaluation of cerebral and spinal cord damage in Amyotrophic Lateral Sclerosis. Neuroimage Clin 14:269-276. Azzarito M, et al. (2020): Tracking the neurodegenerative gradient after spinal cord injury. Neuroimage Clin 26.

[0155] Barry RL, et al. (2022): Selective atrophy of the cervical enlargement in whole spinal cord MRI of amyotrophic lateral sclerosis. Neuroimage Clin 36.

[0156] Bautin P, Cohen- Adad J (2021): Minimum detectable spinal cord atrophy with automatic segmentation: Investigations using an open-access dataset of healthy participants. Neuroimage Clin 32.

[0157] Bischof A, et al. (2022): Spinal Cord Atrophy Predicts Progressive Disease in Relapsing Multiple Sclerosis. Ann Neurol 91 :268-281.

[0158] Van Broeckhoven J, Sommer D, Dooley D, Hendrix S, Franssen AJPM (2021): Macrophage phagocytosis after spinal cord injury: When friends become foes. Brain. Oxford University Press.

[0159] Cohen- Adad J, et al. (2021a): Open-access quantitative MRI data of the spinal cord and reproducibility across participants, sites and manufacturers. Sci Data 8.

[0160] Cohen- Adad J, et al. Xu J (2021b): Generic acquisition protocol for quantitative MRI of the spinal cord. Nature Protocols. Nature Research.

[0161] Cohen AB, et al. (2012): The Relationships among MRI-Defined Spinal Cord Involvement, Brain Involvement, and Disability in Multiple Sclerosis. Journal of Neuroimaging 22: 122- 128.

[0162] Ganzetti et al. (2016): Intensity Inhomogeneity Correction of Structural MR Images: A Data- Driven Approach to Define Input Algorithm Parameters. Front. Neuroinform.10: 10.

[0163] Grolez G, et al. (2018): MRI of the cervical spinal cord predicts respiratory dysfunction in ALS. Sci Rep 8.

[0164] Gros C, et al. (2019): Automatic segmentation of the spinal cord and intramedullary multiple sclerosis lesions with convolutional neural networks. Neuroimage 184:901-915.

[0165] Gudbjartsson H, Patz S (1995): The Rician Distribution of Noisy MRI Data. Magn Reson Med.

[0166] Horsfield MA, et al. (2010): Rapid semi-automatic segmentation of the spinal cord from magnetic resonance images: Application in multiple sclerosis. Neuroimage 50:446-455.

[0167] De Leener B, et al. (2018): PAM50: Unbiased multimodal template of the brainstem and spinal cord aligned with the ICBM152 space. Neuroimage 165: 170-179.

[0168] De Leener B, et al. (2014): Robust, accurate and fast automatic segmentation of the spinal cord. Neuroimage 98:528-536.

[0169] De Leener B, et al. (2017): SCT: Spinal Cord Toolbox, an open-source software for processing spinal cord MRI data. Neuroimage 145:24-43.

[0170] Lundell H, , et al. (2011): Independent spinal cord atrophy measures correlate to motor and sensory deficits in individuals with spinal cord injury. Spinal Cord 49:70-75. El Mendili MM, , et al. (2019): Spinal cord imaging in amyotrophic lateral sclerosis: Historical concepts — novel techniques. Frontiers in Neurology. Frontiers Media S.A.

[0171] Mina Y, et al. (2021): Cervical and thoracic cord atrophy in multiple sclerosis phenotypes: Quantification and correlation with clinical disability. Neuroimage Clin 30.

[0172] Ontaneda D, et al. (2023): Identification and management of subclinical disease activity in early multiple sclerosis: a review. Journal of Neurology. Springer Science and Business Media Deutschland GmbH.

[0173] Prados F, et al. (2020): Generalised boundary shift integral for longitudinal assessment of spinal cord atrophy. Neuroimage 209.

[0174] Rocca MA, et al. (2022): Spinal cord lesions and brain grey matter atrophy independently predict clinical worsening in definite multiple sclerosis: a 5-year, multicentre study. J Neurol Neurosurg Psychiatry 94: 10-18.

[0175] Rocca MA, et al. (2019): Clinically relevant cranio-caudal patterns of cervical cord atrophy evolution in MS. Neurology 93:E1852-E1866.

[0176] Ruggieri S, et al. (2015): Association of deep gray matter damage with cortical and spinal cord degeneration in primary progressive multiple sclerosis. JAMA Neurol 72: 1466-1474. Trolle C, et al. (2023): Spinal cord atrophy after spinal cord injury - A systematic review and meta-analysis. NeuroImage: Clinical. Elsevier Inc.

[0177] Tsagkas C, et al. (2018): Spinal cord volume loss: A marker of disease progression in multiple sclerosis. Neurology 91 :e349-e358.

[0178] Ullmann E, et al. J (2014): Automatic Labeling of Vertebral Levels Using a Robust Template-Based Approach. Int J Biomed Imaging 2014.

[0179] Valsasina P, et al. (2022): Improved Assessment of Longitudinal Spinal Cord Atrophy in Multiple Sclerosis Using a Registration-Based Approach: Relevance for Clinical Studies. Journal of Magnetic Resonance Imaging 55: 1559-1568.

[0180] Wendebourg MJ, et al. (2024): Cervical and thoracic spinal cord gray matter atrophy is associated with disability in patients with amyotrophic lateral sclerosis. Eur J Neurol. https: / / onlinelibrary.wiley.com / doi / 10. I l l 1 / ene.16268.

[0181] Zeydan B, et al. (2018): Cervical spinal cord atrophy: An early marker of progressive MS onset. Neurol Neuroimmunol Neuroinflamm 5.

Claims

Claims1. A computer-implemented method for generating an image of a spinal cord of a subject, said image displaying spinal cord atrophy, comprising: a) providing a reference image of a spinal cord based on pre-recorded MRI data of said subject, wherein image regions of the reference image relating to the spinal cord are pre-identified; b) generating at least one synthetic image with at least one pre-defined atrophy rate of the spinal cord, wherein the generation of said synthetic images comprises the following steps: bl) stripping at least the image regions from the reference image relating to the spinal cord, thereby generating at least one stripped spinal cord region in the reference image and a cord image; b2) filling said stripped spinal cord region with non-zero pixel values relating to cerebrospinal fluid encircling the image regions of the reference image relating to the spinal cord; b3) scaling the cord image by a pre-defined factor relating to a pre-defined atrophy rate; b4) superimposing the scaled cord image onto the filled spinal cord region; c) retrieving said at least one synthetic image with at least one atrophy rate of the spinal cord.

2. The method according to claim 1, wherein spine curvature in the reference image of the spinal cord is removed, typically, wherein the spine curvature is re-introduced into the at least one synthetic image based on the pre-recorded MRI data of said subject.

3. The method according to any of the preceding claims, wherein step a) comprises retrieving said reference image from any of the following: a database, a cloud, a computer, a data storage device; or wherein step a) comprises generating said reference image by an image classification algorithm.

4. The method according to the preceding claim, wherein said image classification algorithm comprises at least the following steps:- identifying the spinal cord region in said pre-recorded MRI data of said subject, typically by partitioning the MR images into different segments comprising sets of voxels assigned to the spinal cord;- registering the identified spinal cord region to a template, typically to a straightened template, for example PAM50 template, thereby generating the reference image;- assigning axial levels to the spinal cord region.

5. The method according to any of the two preceding claims, wherein the image classification algorithm uses at least one classification algorithm selected from the group consisting of at least one convolutional neural network (CNN) technology, deconvolu- tional neural network (DNN) technology, decision trees, point vector, random forest, K-Nearest-Neighbor.

6. The method according to any of the preceding claims, wherein said stripping in step bl) comprises stripping the image regions relating to the spinal cord from the reference image, or wherein step bl) comprises identifying further image regions relating to a dilated spinal cord, and stripping the image regions relating to the spinal cord and the further image regions relating to the dilated spinal cord from the reference image.

7. The method according to any of the preceding claims, wherein the scaling of the cord image is isotropic.

8. The method according to any of the preceding claims, further comprising at least one of the following steps: displaying said retrieved synthetic image on a screen; evaluating said retrieved synthetic image using at least one evaluation operation; and / or transferring said retrieved synthetic image to at least one data storage device.

9. The method according to any of the preceding claims, wherein filling said stripped spinal cord region with non-zero pixel values comprises using at least one mathematical algorithm, wherein the mathematical algorithm comprises one or more of: using at least one filling function.

10. The method according to any of the preceding claims, further comprising step b5) of repeating steps b3) and b4) for different atrophy rates thereby generating a plurality of synthetic images with different atrophy rates of the spinal cord.

11. The method according to any of the preceding claims, further comprising a step of introducing noise, typically by introducing noise based on a Rician distribution and / or on intensity inhomogeneities; a step of assessing spinal atrophy based on said synthetic images; and / or a step of monitoring atrophy rates in a subject based on said synthetic images.

12. The method according to any of the preceding claims, further comprising using said synthetic images in one of the following applications:- diagnosis of spinal atrophy;- prognosis of spinal atrophy;- monitoring spinal atrophy;- training of an image classification algorithm for identifying spinal atrophy;- quantifying the level of atrophy in a clinical setting;- image classification; and / or- image segmentation.

13. A method for training an image classification algorithm using images with at least one atrophy rate of the spinal cord comprising, generating one or more images of a spinal cord of a subject as defined in the method of any of the preceding claims 1 to 12, said one or more images displaying spinal cord atrophy; training the image classification algorithm using said one or more images, or using a plurality of said images, as a training dataset.

14. A method for assessing spinal atrophy comprising generating an image of a spinal cord of a subject displaying spinal cord atrophy by performing the method of any of the preceding claims 1 to 12, further comprising comparing said synthetic image orplurality of synthetic images of the spinal cord to a reference, typically to an earlier set of synthetic images or to a reference MRI dataset; and more typically determining the atrophy rate based on said comparison.

15. A method for monitoring an atrophy rate of a spinal cord, typically in a subject suspected or known to suffer from an atrophy causing disease or disorder, comprising generating a set of synthetic images of a spinal cord of a subject, said images displaying spinal cord atrophy, wherein the images are generated by performing the method as defined in any of the preceding claims 1 to 13, further comprising compar- ing said synthetic images of the spinal cord to a reference, typically to an earlier set of synthetic images or to a reference MRI dataset; and more typically determining a change in the atrophy rate based on said comparison.

16. A computer system or computer network, for performing the method according to any of claims 1 to 12.

Images