Systems and methods for determining risk of acquiring or developing steatohepatitis and / or its complications - Patents.com

Abstract

The present invention relates to a computer-implemented method for determining a subject's risk of suffering from or developing steatohepatitis and / or complications thereof, comprising the steps of receiving (11) at least one image (20) of the subject and at least one clinical and / or biological data (21), both of which have been previously acquired from the subject, calculating (12) a fatty infiltration heterogeneity score on the at least one image based on at least one radiomics feature obtained for a region of interest (ROI) comprising at least one skeletal muscle, combining (13) the calculated fatty infiltration heterogeneity score with the at least one clinical and / or biological data previously acquired from the subject, and determining (14) the subject's risk of suffering from or developing steatohepatitis and / or complications thereof based on a combination of the calculated fatty infiltration heterogeneity score and the at least one clinical and / or biological data (21).

Description

[Technical field]

[0001] The present invention relates to the field of diagnosis in hepatology, more precisely to the diagnosis of steatohepatitis and / or its complications in a subject. In particular, the present invention relates to a device and a method for determining the risk of suffering from or developing steatohepatitis through imaging. [Background technology]

[0002] Steatohepatitis is a rapidly progressive fatty liver disease that is specifically part of the non-alcoholic fatty liver disease (NAFLD) spectrum. Steatohepatitis, particularly non-alcoholic steatohepatitis (NASH), has the unique property of progressing towards liver fibrosis, which can cause progressive liver scarring leading to cirrhosis or towards hepatocellular carcinoma (liver cancer).

[0003] NAFLD is characterized by the presence of hepatic steatosis, i.e., accumulation of lipids in hepatocytes, which develops in the absence of the use of lipogenic drugs, genetic diseases, or inborn errors of metabolism. NAFLD consists of two pathological subtypes: fatty liver disease (NAFL) and NASH. In NAFL, fatty liver disease is characterized by the presence of fat in the liver, but little or no inflammation or liver damage. In NASH, inflammation with hepatic steatosis and hepatocyte damage (balloon degeneration) is present in addition to hepatic steatosis, with or without liver fibrosis (Chalasani N et al., Gastroenterology 2012;142:1592-609).

[0004] Liver cancer is the most common type of primary liver cancer in adults and the second leading cause of cancer-related deaths worldwide. Hepatocellular carcinoma (HCC) accounts for 70-85% of all liver cancers. NAFLD is the most prevalent chronic liver disease in the world and is a proven risk factor for HCC development. The prevalence of NAFLD-associated HCC is alarmingly increasing. A study conducted in Newcastle (UK) reported a more than 10-fold increase in NAFLD-associated HCC between 2000 and 2010. By 2030, the incidence of HCC is predicted to increase by 137% in the United States, with approximately 20% of these new cases being related to NAFLD. Therefore, tools to assess the risk of HCC development in NAFLD patients are urgently needed to optimize monitoring of this large at-risk population.

[0005] NAFLD, NASH, and HCC are generally diagnosed through histological examination of liver samples taken by biopsy by liver pathologists.Liver biopsy is still considered the gold standard for evaluating the presence and / or severity of liver disease in subjects, but it has limitations, especially due to low inter- and intra-observer reproducibility and possible sampling bias due to small sample size.In addition, liver biopsy is an invasive medical procedure, and therefore still involves risk of complications and significant costs. Summary of the Invention [Problem to be solved by the invention]

[0006] Therefore, there is still a need for a non-invasive method for specifically and accurately assessing the presence of steatohepatitis, particularly NASH, and / or its complications in a subject.In particular, such a non-invasive method will be a convenient and practical method for assessing the presence of steatohepatitis, particularly NASH, and / or its complications in a subject, for monitoring the development of NASH complications in patients suffering from NAFLD or NASH, and for monitoring the treatment response of the subject undergoing treatment for NASH. [Means for solving the problem]

[0007] The present invention relates to a computer-implemented method for determining a subject's risk of acquiring or developing steatohepatitis and / or complications thereof, comprising: - receiving at least one image of a subject pre-acquired with a medical imaging technique and at least one clinical and / or biological data pre-acquired from the subject; - calculating a fatty infiltration heterogeneity score on at least one image, said fatty infiltration score being calculated based on at least one radiomics feature obtained for a region of interest (ROI) defined on at least one image to include at least one skeletal muscle, said at least one radiomics feature being energy; - combining the calculated fatty infiltration heterogeneity score with at least one clinical and / or biological data previously obtained from the subject; - determining the subject's risk of suffering from or developing steatohepatitis and / or its complications based on the calculated fatty infiltration heterogeneity score in combination with at least one clinical and / or biological data; The present invention relates to a method comprising the steps of:

[0008] According to one embodiment, the steatohepatitis is non-alcoholic steatohepatitis (NASH).

[0009] According to one embodiment, the step of combining the calculated fatty infiltration heterogeneity score with at least one clinical and / or biological data of the subject comprises using a machine learning model.

[0010] According to one embodiment, the fatty infiltration score is calculated based on the energy and at least one other radiomic feature.

[0011] According to one embodiment, the at least one other radiomic feature is a first order grey level statistic obtained from the image pixels or at least one region of the image.

[0012] According to one embodiment, the at least one other radiomics feature is selected from mean absolute deviation, root mean square, uniformity, minimum intensity, maximum intensity, mean intensity, median, intensity range, intensity variance, intensity standard deviation, skewness, kurtosis, variance, and entropy.

[0013] According to one embodiment, the at least one other radiomics feature is a second order grey level statistic derived from a co-occurrence matrix of the image, said second order grey level statistic being one of: contrast, correlation between adjacent image regions or pixels, energy, uniformity of a first type, uniformity of a second type, inverse difference moments, Sum average, Sum variance, Sum entropy, autocorrelation, Cluster prominence, Cluster shade, Cluster tendency, dissimilarity, normalized inverse difference moments, normalized inverse difference, inverse variance; Run-length grey level statistics, Short run emphasis, Long run emphasis, Run percentage, Gray-level non-uniformity, Run length non-uniformity, Low gray level run emphasis, High gray level run emphasis, shape and size based features such as circumference, cross-sectional area, major axis length, maximum diameter and volume; Small area emphasis, Large area emphasis, Intensity variability, Size-zone variability, Zone percentage, Low intensity emphasis, High intensity emphasis, Low intensity small area emphasis, High intensity small area emphasis, The emphasis is selected from size-zone matrix based features of grey levels such as emphasis, Low intensity large area emphasis, and High intensity large area emphasis.In one embodiment, the secondary gray-level statistics include various feature subgroups such as Gray-Level Cooccurrence Matrix (GLCM), Gray-Level Run-length Matrix (GLRLM), Gray-Level Size Zone Matrix (GLSZM), and Gray-Level Distance Zone Matrix (GLDZM), Neighborhood Gray-Tone Difference Matrix (NGTDM), or Neighborhood Gray-Level Dependence Matrix (NGLDM).

[0014] According to one embodiment, the fatty infiltration heterogeneity score is a function of the energy and at least one other radiomic feature, each weighted by a weighting factor.

[0015] According to one embodiment, the at least one clinical and / or biological data is selected from age, sex, and personal or family history of metabolic-related adverse events.

[0016] According to one embodiment, the machine learning model is a random forest.

[0017] According to one embodiment, the complication of steatohepatitis is selected from hepatocellular carcinoma (HCC), fibrosis (eg, liver fibrosis), cirrhosis, and cardiovascular disease.

[0018] According to one embodiment, the subject suffers from a dysmetabolic condition, obesity, metabolic syndrome, overweight, and / or fatty liver disease, in particular non-alcoholic fatty liver disease (NAFLD).

[0019] According to one embodiment, the subject has NASH.

[0020] According to one embodiment, the medical imaging technique is computed tomography, magnetic resonance imaging, or ultrasound.

[0021] The present invention further provides a device for determining a subject's risk of acquiring or developing non-alcoholic steatohepatitis (NASH) and / or complications thereof, comprising: - at least one input adapted to receive at least one image of the subject pre-acquired with a medical imaging technique and at least one clinical and / or biological data pre-acquired from the subject; - at least one processor, calculating a fatty infiltration heterogeneity score on at least one image, the fatty infiltration score being calculated based on at least one radiomics feature obtained for a region of interest defined on the at least one image to include at least one skeletal muscle, the at least one radiomics feature being energy; combining the calculated fatty infiltration heterogeneity score with at least one clinical and / or biological data previously obtained from the subject; determining the subject's risk of suffering from or developing NASH and / or its complications based on the calculated fatty infiltration heterogeneity score in combination with at least one clinical and / or biological data; at least one processor configured to - at least one output adapted to provide said risk; The present invention relates to a device comprising:

[0022] The device for determining risk can be implemented in various ways, in particular as stand-alone software installed on a computer, a cloud-based service, or an application programming interface (API), together or separately, with a suitable associated user interface.

[0023] 2. A computer program comprising software code adapted to, when executed by a processor, perform a method for predicting a subject's risk of suffering from or developing steatohepatitis (NASH) and / or its complications according to any one of the above embodiments.

[0024] Furthermore, the present disclosure relates to a computer program comprising software code adapted to execute the method for determining a risk according to any of the above modes of execution when said program is executed by a processor.

[0025] The present disclosure further relates to a non-transitory computer readable program storage device tangibly embodying a program of instructions executable by a computer to perform a method for determining risk in accordance with the present disclosure.

[0026] 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 above. More specific examples are provided below: portable computer diskettes, hard disks, ROMs, EPROMs (Erasable Programmable ROMs) or flash memories, portable CD-ROMs (Compact-Disc ROMs), but it should be understood that these are merely illustrative and not exhaustive lists, as would be readily apparent to one skilled in the art.

[0027] definition In the present invention, the following terms have the following meanings:

[0028] The terms "adapted" and "configured" are used in this disclosure to similarly broadly encompass the initial configuration of the device, subsequent adaptation or supplementation, or any combination thereof, whether effected through material or software means (including firmware).

[0029] The term "processor" should not be construed as being limited to hardware capable of executing software, but generally refers to a processing device that may include, for example, a computer, a microprocessor, an integrated circuit, or a programmable logic device (PLD). A processor may also encompass one or more graphics processing units (GPUs), whether utilized for computer graphics and image processing or other functions. Furthermore, instructions and / or data enabling the execution of the relevant functions and / or resulting functions may be stored on 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). Instructions may be stored in hardware, software, firmware, or any combination thereof, among others.

[0030] An "image" refers to a visual representation of a scene currently including at least a portion of a subject including at least one skeletal muscle at a given time. An image can therefore be a frame or a set of frames.

[0031] "Image segmentation" consists of partitioning a digital image into segments by assigning labels to image pixels with the goal of facilitating subsequent processing.

[0032] "Machine learning (ML)" refers to computer algorithms that improve automatically through experience in a traditional manner based on training data that allows for tuning of parameters of a computer model through the reduction of the gap between expected outputs extracted from the training data and the evaluated outputs calculated by the computer model.

[0033] A "dataset" is a collection of data used to build an ML mathematical model to make data-driven predictions or decisions. In "supervised learning" (i.e., inferring a function from known input-output examples in the form of labeled training data), three types of ML datasets (also called ML sets) are typically dedicated to three kinds of tasks, respectively: "training", i.e., fitting parameters; "validation", i.e., adjusting ML hyperparameters (parameters used to control the learning process); and "testing", i.e., checking whether the latter model provides satisfactory results independently of the training dataset used to build the mathematical model.

[0034] "Radiomics features" refer to features extracted from radiological medical images using data characterization algorithms. The concept of radiomics is based on the premise that biomedical images contain information of disease-specific processes that are not perceptible to the human eye and therefore cannot be obtained through conventional visual inspection of the generated images. Through mathematical extraction of the spatial distribution of signal intensity and pixel interrelationships, radiomics quantifies textural information. Furthermore, visually perceptible differences in the intensity, shape or texture of an image can be quantified by radiomics, thus overcoming the subjective nature of image interpretation. Radiomics therefore does not imply the automation of the diagnostic process, but rather provides additional data to the existing diagnostic process. Radiomic or intensity-based statistical features describe how intensities are distributed within a region of interest (ROI). N v The set of intensities for voxels is X gl ={X gl,1 ,X gl,2 ,...,X gl,Nv The intensity histogram is expressed as the original intensity distribution X gl can be generated by discretizing X into intensity bins. d ={Xd,1 ,X d,2 ,...,X d,Nv} in the ROI intensity mask v N of voxels g Let H be the set of discretized intensities. 1 ,n 2 ,...,n Ng} to X d The frequency count n of each discretized intensity i in i Then, for each discretized intensity i, the probability of occurrence p i p i =n i / N v The mean μ and standard deviation σ of the grey levels of the voxels assigned to the ROI are calculated. For example, the "energy" is

number

number

number

number

number

[0035] "Biological data" refers to parameters that can be measured in a sample taken from a subject. For example, the parameters can be measured in a blood, plasma, or serum sample taken from a subject by commonly used clinical tests. Blood sugar (also called blood glucose or glycemia), cholesterol, high density lipoprotein (HDL), low density lipoprotein (LDL), alanine aminotransferase (ALT), and aspartate aminotransferase (AST) are non-limiting examples of biological data.

[0036] "Clinical data" refers to data collected from external observation of a subject without the use of clinical testing. Age, race, sex, diastolic blood pressure (DBP) and systolic blood pressure (SBP), family history (FamHX), height (HT), weight (WT), waist and hip circumference, and body mass index (BMI) are non-limiting examples of clinical data.

[0037] "AUROC" stands for area under the ROC curve, a measure of the accuracy of a diagnostic test. In statistics, a receiver operating characteristic (ROC), or ROC curve, is a graphical plot that shows the performance of a binary classifier system as its discrimination threshold is varied. The curve is constructed by plotting sensitivity against specificity on a continuum from 0 to 1 (usually a specificity of 1). ROC curves and AUROC are well known in the field of statistics (i.e., they show how well a model can distinguish between classes).

[0038] "NAFLD" (non-alcoholic fatty liver disease) is characterized by the presence of hepatic steatosis, which develops in the absence of heavy alcohol consumption, use of lipogenic drugs, genetic diseases or inborn errors of metabolism. NAFLD includes various severity stages of the disease, the less severe NAFL (non-alcoholic fatty liver) and the more severe NASH (non-alcoholic steatohepatitis).

[0039] "NASH" (nonalcoholic steatohepatitis) refers to the presence of a score ≧1 for each of the three components of NAS: steatosis, intralobular inflammation, and ballooning, with or without fibrosis.

[0040] "Risk" in the context of the present invention relates to the probability of an event occurring over a certain period (e.g., 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, or 12 months, or 2-7 years), and may refer to the "absolute" or "relative" risk of a subject. Absolute risk can be measured with reference to actual observations after measurements of the relevant time cohort, or with reference to index values ​​derived from statistically valid historical cohorts that have been followed over the relevant period. Relative risk refers to the ratio of the absolute risk of a subject compared to the absolute risk of a low-risk cohort or the average population risk, and may vary depending on how clinical risk factors are evaluated.

[0041] "Score" refers to any numerical value obtained by a mathematical combination of markers. In one embodiment, the score is a bounded numerical value obtained by a mathematical function. In one embodiment, the score may range from 0 to 1.

[0042] The "sensitivity" of a diagnostic test refers to the proportion of patients who have the diagnostic target who are properly identified as such (i.e., the percentage of patients who have the diagnostic target who are properly identified as positive by the diagnostic test).

[0043] The "specificity" of a diagnostic test represents the proportion of patients who do not have the diagnostic target who are properly identified as such (i.e., the percentage of patients who do not have the diagnostic target who are properly identified as negative by the diagnostic test).

[0044] "Steatosis" refers to the accumulation of fat in the liver as lipid droplets within hepatocytes.

[0045] "Steatohepatitis" refers to fatty liver disease associated with metabolic dysfunction, characterized by liver inflammation, liver cell damage, and accumulation of fat in the liver.

[0046] "Subject" refers to a mammal, preferably a human.

[0047] "Skeletal muscle" refers to a voluntary muscle under the control of the somatic nervous system.

[0048] "Complication" or "complication thereof" refers to the adverse events associated with steatohepatitis, particularly NASH, or the progression of steatohepatitis, particularly NASH, in a subject.Complications of steatohepatitis include, for example, fibrosis (e.g., liver fibrosis), liver cirrhosis, hepatocellular carcinoma (HCC), and cardiovascular disease.

[0049] "Energy" is a measure of the magnitude of the voxel values ​​in an image. It is a measure of the uniformity of the image array. Larger values ​​mean that the sum of the squares of these values ​​is larger, and thus reflect a higher uniformity, or a smaller range of discrete intensity values. This feature makes volume a confounding variable.

[0050] "Entropy" describes the uncertainty / randomness of image values. It describes the average amount of information needed to encode an image value.

[0051] "Skewness" describes the asymmetry of the distribution of values ​​around the mean. Depending on how far the tails extend and where most of the distribution is concentrated, this value can be positive or negative.

[0052] "Kurtosis" is a measure of the "peakedness" of the distribution of values ​​in an image ROI. Higher kurtosis means that most of the distribution is concentrated towards the tail rather than the mean. Conversely, lower kurtosis means that most of the distribution is concentrated towards a spike closer to the mean.

[0053] "Variance" is the average of the squared distances of each intensity value from the mean. It is a measure of the spread of the distribution around the mean.

[0054] "Fibrosis" in the present invention particularly refers to hepatic fibrosis, which is a pathological lesion of the liver consisting of scar tissue containing fibrous proteins or glycoproteins (collagens, proteoglycans...).

[0055] "NAFLD Activity Score (NAS)" refers to a system for scoring histological features of nonalcoholic fatty liver disease (NAFLD). NAS ranges from 0 to 8 and corresponds to the sum of the scores for steatosis, intralobular inflammation, and ballooning. The NAS scoring system is commonly used in the histological diagnosis of NASH and is defined as the presence of a score ≧1 for each of the three components of NAS. In one embodiment, an NAS score ≧4 defines active NASH. In one embodiment, an NAS score <3 defines inactive NASH.

[0056] The "NAFLD fibrosis score" is an algorithm that estimates the amount of scarring in the liver. The algorithm is based on six clinical and laboratory tests: age, body mass index, presence of fasting hypoglycemia or diabetes, aspartate aminotransferase (AST) to alanine aminotransferase (ALT) ratio, platelet count, and serum albumin concentration.

[0057] "Histological Inflammation Score" is a score based on the grade of inflammation seen on histological examination of a liver specimen (i.e., liver biopsy), characterized according to severity (none, mild, moderate, severe), which refers to the number of foci of inflammatory cells in a microscopic high-power field, and their location within the hepatic lobule. [Brief description of the drawings]

[0058] [Figure 1] FIG. 1 is a block diagram that generally represents a particular mode of a device for predicting the risk of developing non-alcoholic steatohepatitis (NASH) and / or its complications in accordance with the present disclosure. [Diagram 2] 2 is a flow chart showing successive steps carried out in the device for predicting of FIG. 1; [Diagram 3] 1 is a table showing demographic and histological characteristics of the NAFLD population. [Figure 4] 1 is a table showing that NAFLD patients with HCC have a higher degree of muscle fat depletion than those without HCC. [Diagram 5] FIG. 1 is a table showing that HCC is a significant predictor of muscle fatty tissue in patients with NAFLD. [Figure 6] 1 is a table showing that myocellular fat transformation is strongly associated with NASH-associated HCC, independent of sex, age, gender, visceral fat, NASH activity, and fibrosis severity. [Figure 7] FIG. 1 depicts a magnetic resonance imaging-proton density fat fraction (MRI-PDFF) virtual muscle biopsy to assess muscle fat density at the third lumbar vertebral level. [Figure 8] FIG. 1 shows that HCC is associated with severe muscle fatty transformation in NASH patients, where fat mass was measured in three muscles showing heterogeneous fatty infiltration. [Figure 9] FIG. 13. Primary radiomics features above mean PDFF for predicting HCC in NASH patients. [Figure 10] 1 is a table showing that HCC prevalence is not explained by fibrosis severity in this NAFLD cohort. [Figure 11] 1 is a table showing that muscle fat transformation is specifically associated with HCC in patients with NAFLD. [Figure 12] 1 is a table showing that increased PDFF of erector spinae is associated with higher risk of HCC in NASH patients. [Figure 13] 1 is a flow chart of patient selection. [Figure 14] 1 is a graph showing that HCC is not associated with increased fatty infiltration in patients with NAFL. [Figure 15] 15a and 15b are graphs illustrating that absolute lipid content is higher in NASH-associated HCC patients than in non-HCC patients. [Figure 16]Figures 16a, 16b, 16c, and 16d are graphs showing that muscle fatty infiltration is more severe in patients with NASH and HCC compared to non-HCC patients, independent of HCC severity. [Figure 17] 17a, 17b, 17c, 17d, and 17e show that NASH is associated with increased fatty infiltration compared to NAFL patients. [Figure 18] 18a, 18b, and 18c are graphs showing that NAFLD fibrosis scores correlate with erector spinae energy, kurtosis, and entropy. [Figure 19] Figures 19a, 19b, 19c, and 19d are graphs showing that NAFLD Activity Score (NAS) correlates with leg strain and histological inflammation score correlates with leg strain and energy and psoas strain. [Figure 20] 1 is a graph showing the correlation between erector spinae muscle dispersion and SCORE2, which indicates the risk of cardiovascular disease. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0059] The present description illustrates the principles of the disclosure, and thus, those skilled in the art will recognize that various arrangements can be devised that, although not explicitly described or shown herein, embody the principles of the disclosure and are included within its scope.

[0060] All examples and conditional language recited in this specification are intended for teaching purposes to aid the reader in understanding the principles of the present disclosure and the concepts the inventors contribute to the advancement of the art, and should not be construed as being limited to such specifically recited examples and conditions.

[0061] Moreover, all statements herein reciting principles, aspects, and embodiments of the disclosure, as well as specific examples thereof, 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.

[0062] 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, any flow charts, flow diagrams, or the like will be understood to represent various processes that may be performed by a computer or processor, whether or not a computer or processor is explicitly depicted.

[0063] 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.

[0064] It should be understood that the elements illustrated 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.

[0065] The present disclosure will be described in relation to a particular functional embodiment of a device 1 for determining a subject's risk of acquiring or developing steatohepatitis, in particular non-alcoholic steatohepatitis (NASH) and / or its complications, as illustrated in FIG.

[0066] The device 1 is adapted to generate a risk prediction 30 for a subject to suffer from or develop steatohepatitis, in particular NASH and / or its complications, based on image data 20 representing fatty infiltration heterogeneity in skeletal muscle and clinical and / or biological data 21 relating to the subject.

[0067] At least one image 20 of the object may be generated using a medical imaging technique such as computed tomography, magnetic resonance imaging, or ultrasound.

[0068] The images may be grayscale or color images. The image data 21 may include numerical data, such as digital data. These data may include, for example, individual image data in a compressed format conforming to the Joint Photographic Experts Group (JPEG), JPEG2000, DICOM, or High Efficiency Image File Format (HEIF) standards, as is well known to those skilled in the art of image compression. The images include individual image data in a compressed format recognized by Digital Imaging and Communication in Medicine (DICOM).

[0069] Thus, the present invention further relates to a method for determining a subject's risk of acquiring or developing steatohepatitis, in particular NASH, and / or its complications, comprising the use of the device 1 described herein.

[0070] In one embodiment, the method is a method for assessing whether a subject is at risk of suffering from or developing steatohepatitis, in particular NASH, and / or complications thereof.

[0071] In one embodiment, the method is a method for diagnosing steatohepatitis, in particular NASH, and / or complications thereof in a subject, i.e. for determining whether said subject suffers from steatohepatitis, in particular NASH, and / or complications thereof.

[0072] In one embodiment, the method is a method for predicting whether a subject is at risk of developing steatohepatitis, particularly NASH, and / or its complications. In one embodiment, the method is for determining the risk of a subject developing steatohepatitis, particularly NASH, and / or its complications in the future, for example, over the next 2, 3, 4, 5, 6, 7, 8, 9, 10, or 11 months, or over the next year, or within the next 2-7 years.

[0073] In one embodiment, the method is for assessing the presence and / or severity of NAFLD. In one embodiment, the severity of NAFLD is assessed by fibrosis score, NAFLD activity score (NAS), and / or histological inflammation score. The fibrosis score estimates liver fibrosis in NAFLD patients. The fibrosis score includes six variables: age, hyperglycemia, body mass index, platelet count, albumin, and AST / ALT ratio. NAS includes the measurement of scores of three parameters: intralobular inflammation, hepatocellular ballooning, and fibrosis. The NAS scoring system is commonly used to assess the severity of NAFLD and to diagnose NASH histologically, and is defined as the presence of a score ≧1 for each of the three components of NAS. In one embodiment, an NAS score ≧4 defines active NASH. In one embodiment, an NAS score <3 defines inactive NASH. In one embodiment, the clinically relevant definition of steatohepatitis is constantly evolving / improving, and in the future may be defined based on patterns detected by machine learning algorithms.

[0074] In one embodiment, the method is for determining the risk of developing steatohepatitis, particularly NASH, in a subject not suffering from steatohepatitis, particularly NASH.

[0075] In one embodiment, the method is for determining the risk of developing a complication of NASH in a subject with uncomplicated NASH.In one embodiment, the method is for determining the risk of developing a complication of NASH in a subject with NASH.

[0076] In one embodiment, the method is for diagnosing NASH and determining the risk of developing complications of NASH in a subject.

[0077] The present invention further relates to a method for providing adapted care to a subject suffering from or at risk of developing steatohepatitis, in particular NASH, and / or complications thereof, comprising: - determining the risk of acquiring or developing steatohepatitis, in particular NASH, and / or its complications in a subject using a method as described herein; - providing tailored care to said subject; The present invention relates to a method comprising the steps of:

[0078] In one embodiment, the subject is at risk of suffering from or developing steatohepatitis, particularly NASH, and / or complications thereof, and appropriate care may include, for example, monitoring the subject for the development of steatohepatitis, particularly NASH, or complications thereof, e.g., by repeating the method for determining the risk of suffering from or developing NASH and / or complications thereof after 0, 1, 2, 3, 4, 5, 6, or 7 months.

[0079] In one embodiment, the subject is at risk of suffering from or developing NASH and / or its complications, and appropriate care may include, for example, administering to the subject a treatment to treat NASH or to treat a complication of NASH.

[0080] According to one embodiment, the method of the present invention comprises receiving clinical and / or biological data 21 and combining them with the calculated fatty infiltration heterogeneity score.

[0081] Clinical and / or biological data 21 also pre-obtained from the subject may include information collected independently of the image data 20, such as information regarding age, sex, biological data, and personal or family history of metabolic-related adverse events. These data may also be obtained from medical databases. Other non-limiting examples of clinical data include, but are not limited to, race, diastolic and systolic blood pressure, height, weight, waist and hip circumference, and body mass index.

[0082] In one embodiment, the method of the invention further comprises the step of comparing the calculated fatty infiltration heterogeneity score in combination with at least one clinical and / or biological data 21 to a reference value.

[0083] In one embodiment, the reference value corresponds to a calculated fatty infiltration heterogeneity score together with at least one clinical and / or biological data 21 measured in a reference population. In one embodiment, the reference value is derived from a population study, for example including subjects having a similar age range or subjects of the same or similar racial group.

[0084] According to one embodiment, the reference value is derived from the calculated fatty infiltration heterogeneity score scale together with at least one clinical data and / or biological data from one or more substantially healthy subjects.In one embodiment, the "substantially healthy subject" is a subject that has not been diagnosed with steatohepatitis, particularly NASH, or its complications.In one embodiment, the "substantially healthy subject" is a subject that has not been diagnosed or identified as having or at risk of developing steatohepatitis, particularly NASH, or its complications.

[0085] According to one embodiment, the reference value is derived from the calculated scale of fatty infiltration heterogeneity score together with at least one clinical and / or biological data from one or more subjects diagnosed with steatohepatitis, particularly NASH, and / or with complications of steatohepatitis, particularly NASH. In one embodiment, the reference value is derived from the calculated scale of fatty infiltration heterogeneity score together with at least one clinical and / or biological data from one or more subjects previously determined to be at risk of developing steatohepatitis, particularly NASH, and / or with complications of steatohepatitis, particularly NASH.

[0086] The device 1 for making predictions is associated with a training device suitable for setting the machine learning model parameters of the device 1 in an appropriate manner as described below.

[0087] While the device 1 and training device described herein are versatile and have several functions that can be performed alternatively or in any cumulative manner, other implementations within the scope of this disclosure include devices having only some of the functions.

[0088] Each of the device 1 and the training devices is advantageously an apparatus or physical piece of apparatus designed, configured and / or adapted to perform the above-mentioned functions and to provide the above-mentioned effects or results. In alternative implementations, the device 1 and any of the training devices are embodied as a set of apparatus or physical pieces of apparatus, whether grouped on the same machine or on different, possibly remote, machines. The device 1 and / or the training devices may, for example, be distributed on a cloud infrastructure and have functionality available to users as a cloud-based service, or have remote functionality accessible through an API.

[0089] The device 1 for prediction and the training device may be integrated into the same device or set of devices and may be intended for the same user. In other implementations, the structure of the training device may be completely independent of the structure of the device 1 and may be provided to other users. For example, the device 1 may have a parameterized model available to the operator for risk prediction, which model is entirely configured from previous trainings performed upstream by other players using the training device.

[0090] In the following, the modules should be understood as functional entities, not as material physically separate components. They can therefore be embodied either grouped together in the same tangible concrete component or distributed across several such components. Also, each of these modules may itself be shared between at least two physical components. Furthermore, the modules are also implemented in hardware, software, firmware or a mixed form thereof. They are preferably embodied in at least one processor of the device 1 and the training device.

[0091] The device 1 comprises a receiving module 41 for receiving at least one image (i.e. image data) 20 of a subject and at least one clinical and / or biological data 21, as well as ML parameters of a machine learning (ML) model. The at least one image 20 of the subject, the at least one clinical and / or biological data 21, and / or the ML parameters may be stored in one or more local or remote databases 10. The latter may take the form of storage resources available from any kind of suitable storage means, which may in particular be a RAM or an EEPROM (Electrically-Erasable Programmable Read-Only Memory), such as a flash memory, possibly in a Solid-State Disk (SSD). In an advantageous embodiment, the ML parameters have been generated beforehand by a system including the device for training. Alternatively, the ML parameters are received from a communication network.

[0092] The device 1 optionally further comprises a pre-processing module (not shown) for pre-processing the received image data 20, possibly clinical data and / or biological data 21. The pre-processing module may in particular be adapted to standardize the received image data 20 for efficient reliable processing. It may for example transform the image data 20 by image decompression. According to various configurations, the pre-processing module is adapted to perform only some or all of the above functions in any manner suitable for the subsequent processing stages and in any possible combination.

[0093] In an advantageous mode, the pre-processing module is configured to pre-process the image data 20 to standardize the images, which may increase the efficiency of downstream processing by the device 1. Such standardization may be particularly useful when the images utilized originate from different sources, possibly including different imaging systems.

[0094] Standardization is advantageously applied in a similar manner to the image data 20 and to the images of the training data set (e.g. by the training device). In particular, the device 1 can handle image data coming from a given type of source and its parameters are obtained from training based on image data obtained with a different type of source. Thanks to the standardization, differences between the sources are eliminated or minimized. This can make the device 1 more efficient and reliable.

[0095] The device 1 also comprises a calculation module 42 for calculating a score representative of fatty infiltration heterogeneity in at least one skeletal muscle of the subject. It can be observed that the operations by the receiving module 41, the pre-processing module and the calculation module 42 are not necessarily sequential in time, but may overlap, parallel or alternate in any suitable manner. For example, new image data may be received and pre-processed successively over time while the calculation module 42 is working with previously acquired image data 20. In an alternative example, a batch of image data 20 corresponding to the acquisition of a complete sequence may be completely received and pre-processed before being sent to the calculation module 42.

[0096] The calculation module 42 may be configured to apply a segmentation algorithm to the image data 20 to segment the at least one skeletal muscle. Alternatively, information regarding the boundary of the at least one skeletal muscle on the data image may be received as an input of the module 42 after being manually drawn by a user. The result of the segmentation or manual drawing is used by the module 42 to define a region of interest (ROI) on the image data 20 that includes pixels of the image data 20 associated with the at least one skeletal muscle.

[0097] The at least one skeletal muscle may be a muscle of the back or a muscle of the lower limbs. The muscles of the lower limbs are particularly observed using ultrasound. In one embodiment, the at least one skeletal muscle is selected from the group consisting of the psoas, quadratus lumborum, erector spinae, obliques, and rectus muscles. In one embodiment, the lower limbs may be muscles of the leg. In one embodiment, the muscles of the leg include a plurality of muscles selected from the group consisting of the gastrocnemius, soleus, and tibialis anterior.

[0098] The calculation module 42 is configured to calculate at least one radiomics feature on the ROI. In one embodiment, the at least one radiomics feature is an energy. In fact, this energy can advantageously represent the (in)homogeneity in the ROI. The calculation module 42 can also calculate at least one other radiomics feature, which is a first-order gray-level statistic obtained from the image pixels or at least one region of the image. Said other radiomics feature, which is a first-order gray-level statistic, can be selected from the mean absolute deviation, the root mean square, uniformity, minimum intensity, maximum intensity, mean intensity, median, intensity range, intensity variance, intensity standard deviation, skewness, kurtosis, variance and entropy. The advantage of using first-order gray-level statistics as radiomics features is that they represent the distribution of the individual voxel values, and therefore the distribution of the fatty infiltration, regardless of the spatial relationship.

[0099] In one embodiment, the at least one other radiomics feature is a second order grey level statistic derived from a co-occurrence matrix of the image, said second order grey level statistic being one of: contrast, correlation between adjacent image regions or pixels, energy, uniformity of a first type, uniformity of a second type, inverse difference moments, Sum average, Sum variance, Sum entropy, autocorrelation, Cluster prominence, Cluster shade, Cluster tendency, dissimilarity, normalized inverse difference moments, normalized inverse difference, inverse variance; Run-length grey level statistics, Short run emphasis, Long run emphasis, Run percentage, Gray-level non-uniformity, Run length non-uniformity, Low gray level run emphasis, High gray level run emphasis, shape and size based features such as circumference, cross-sectional area, major axis length, maximum diameter and volume; Small area emphasis, Large area emphasis, Intensity variability, Size-zone variability, Zone percentage, Low intensity emphasis, High intensity emphasis, Low intensity small area emphasis, High intensity small area emphasis, The features are selected from size-zone matrix-based features of grey levels such as Low intensity large area emphasis, Low intensity large area emphasis, and High intensity large area emphasis. The advantage of using second-order grey level statistics as radiomics features is that they provide a measure of the spatial arrangement of voxel intensities in parallel with intramuscular heterogeneity, further refining the characterization of fat infiltration patterns within skeletal muscle.

[0100] The calculation module 42 is also configured to use the at least one radiomics feature to calculate a score representative of the fat infiltration heterogeneity in the at least one skeletal muscle under investigation. The fat infiltration heterogeneity score may be a function of the energy and at least one other radiomics feature, each weighted by a weighting factor. The weighting factors may be pre-derived and stored in the database 10.

[0101] The score representing the fatty infiltration heterogeneity in at least one skeletal muscle under investigation can be calculated in various manners depending on the risk to be determined. In one example, to estimate the risk of steatohepatitis, the fatty infiltration heterogeneity score can be obtained as a weighted sum of energy and skewness, to estimate the risk of HCC (i.e., one complication), the fatty infiltration heterogeneity score can be obtained as a weighted sum of energy and entropy, and to estimate the risk of cardiovascular disease (i.e., one complication), the fatty infiltration heterogeneity score can be obtained as a weighted sum of energy and variance. In this example, a weighting factor is adapted to each radiomics feature and each calculated risk. It is known to those skilled in the art how to select the relevant radiomics feature for a given prediction task and adapt the weighting factor according to the radiomics feature.

[0102] The device 1 also comprises a combination module 43 configured to combine the calculated fatty infiltration heterogeneity score with at least one clinical and / or biological data 21 previously obtained from the subject. Said combination can be obtained using a weighted sum, where the clinical and / or biological data 21 are weighted with predefined weighting factors, respectively, and summed to the calculated fatty infiltration heterogeneity score. The at least one clinical and / or biological data 21 can be the sex and age of the subject, or any other clinical or anthropometric parameter. A person skilled in the art can select the relevant clinical and / or biological data 21 for a given prediction task and adapt the predefined weighting factors according to said clinical and / or biological data 21. Said combination can be performed using a machine learning model that receives the fatty infiltration heterogeneity score and the clinical and / or biological data 21 as input. The machine learning model can be selected from predictive modeling techniques such as decision trees or support vector machines. In an advantageous embodiment, random forests or support vector machines or neural networks are used as machine learning models, since they are all the best performing models in disease risk prediction. The machine learning model can be trained in a supervised manner on the training device and its performance can be validated using k-fold cross-validation techniques. The machine learning parameters determined during training on the training device can be stored in the database 10.

[0103] The device 1 comprises a determination module 44 configured to determine the subject's risk 30 of suffering from or developing steatohepatitis, in particular NASH and / or its complications, based on the output of the combination module 43, which combines the calculated fatty infiltration heterogeneity score with at least one clinical and / or biological data 21. The module 44 can also output the risk 30.

[0104] Module 44 can provide a more accurate risk 30 as long as sufficient information is available on the image data and clinical and / or biological data, which may take the visual form of a written string or a point on a graph indicating the degree of probability that the subject has or will develop steatohepatitis, particularly NASH, or its complications.

[0105] The device 1 interacts with a user interface 11, through which a user can input or retrieve information. The user interface 11 comprises any means suitable for inputting or retrieving data, information or instructions, in particular visual, tactile and / or audio capabilities which may include any or some of the following means well known to those skilled in the art: screen, keyboard, trackball, touchpad, touchscreen, loudspeaker, voice recognition system.

[0106] According to one embodiment, the subject is male, hi another embodiment, the subject is female.

[0107] According to one embodiment, the subject is an adult. According to the present invention, an adult is a subject that is older than 18, 19, 20, or 21 years old. In another embodiment, the subject is a child. According to the present invention, a child is a subject that is younger than 21, 20, 19, or 18 years old, such as 17, 16, 15, 14, 13, 12, 11, 10, or 9 years old.

[0108] According to one embodiment, the subject is a substantially healthy subject, meaning that the subject has not been previously diagnosed or identified as suffering from or suffering from NAFLD, NAFL, or NASH.

[0109] According to one embodiment, the subject is a patient who is waiting for or receiving medical care, or has been, is, or will be the subject of medical treatment, or is monitored for the onset or progression of a disease, i.e. a warm-blooded animal, more preferably a human. According to one embodiment, the subject is a human patient who is being treated and / or monitored for the onset or progression of a liver disease or condition (preferably NAFLD, NAFL, or NASH).

[0110] According to one embodiment, the subject suffers from a dysmetabolic condition, obesity, metabolic syndrome, overweight, and / or fatty liver disease (particularly NAFLD). In one embodiment, the subject suffers from obesity, metabolic syndrome, overweight, and / or fatty liver disease (particularly NAFLD). According to one embodiment, the subject suffers from steatohepatitis. In an embodiment, the subject suffers from NASH. In one embodiment, the method of the invention is for determining the risk of acquiring or developing complications of steatohepatitis, particularly NASH, in a subject diagnosed with steatohepatitis, particularly NASH.

[0111] According to one embodiment, the complication of steatohepatitis, particularly NASH, is selected from hepatocellular carcinoma (HCC), fibrosis (e.g., liver fibrosis), liver cirrhosis, and cardiovascular disease.In one embodiment, the cardiovascular disease is selected from atherosclerosis, coronary heart disease, cerebrovascular disease, cardiomyopathy, heart failure, valvular heart disease, and cardiac arrhythmia.Therefore, in one embodiment, the method of the present invention is for determining the risk of suffering from inflammation (e.g., liver inflammation), fibrosis (e.g., liver fibrosis), or liver cirrhosis, or HCC in subjects diagnosed with NASH.

[0112] In its automatic operation, the device 1 may for example carry out the following process (FIG. 2): receiving image data 20 and clinical and / or biological data 21, both of which have been previously acquired from a subject (step 11); calculating a fatty infiltration heterogeneity score on at least one image, said fatty infiltration score being calculated based on at least one radiomics feature obtained for a region of interest (ROI) defined on the at least one image to include at least one skeletal muscle (step 12); combining the calculated fatty infiltration heterogeneity score with at least one clinical and / or biological data 21 previously obtained from the subject (step 13), - Determining and providing the subject's risk 30 of suffering from or developing steatohepatitis, particularly NASH, and / or complications thereof based on the calculated fatty infiltration heterogeneity score in combination with at least one clinical and / or biological data (step 14).

[0113] A specific apparatus may embody the above-mentioned device 1 as well as a training device, which may for example correspond to a workstation, a laptop, a tablet, a smartphone or a head-mounted display (HMD).

[0114] Working Example The present invention is further illustrated in the following examples. Example 1: Materials and Methods

[0115] This retrospective study (project number 2017-030) was approved by the local IRB. Due to the nature of the retrospective study, informed consent was waived. Between January 2014 and June 2017, 196 patients who underwent liver biopsy or hepatectomy and had a liver MRI with a multi-echo gradient-echo acquisition sequence within 4 months were included in the study. First, patients with confounding conditions for hepatic steatosis (i.e., alcohol intake, chemotherapy, viral infection, etc.) were excluded. Patients with adipocyte fibrosis (i.e., histologically defined as a steatosis score of at least 1, an activity score of 0, and ≥ F3) or inadequate liver biopsy on MRI were excluded. Finally, acquisition data from 4 patients were excluded because they were used for algorithm adjustment, leaving a total of 72 patients for analysis (Figure 3).

[0116] Liver histology was based on liver biopsy (n = 60) or hepatectomy (n = 12). Eligibility for liver biopsy / resection was as follows: 25 patients with NAFLD / NASH evaluation, 47 patients with focal liver lesions, including 15 patients with benign lesions (5 patients with hepatocellular adenoma and 10 patients with focal nodular hyperplasia) and 30 patients with malignant lesions (20 patients with HCC, 1 patient with cholangiocarcinoma, and 8 patients with liver metastases). For each patient, an experienced pathologist (VP) classified hepatic steatosis, activity, and fibrosis according to the SAF score. Steatosis was assessed by the % of hepatocytes containing intracytoplasmic lipid droplets from S0 to S3, S0: <5%, S1: 5%-33%, S2: 34%-66%, S3: >66%. Activity (A0-A4) was calculated by the addition of balloon degeneration and intralobular inflammation grades. Liver fibrosis was classified as absent (F0), mild (F1, perisinusoidal or periportal), moderate (F2, periportal and perisinusoidal), F3 (bridging fibrosis), and F4 (cirrhosis). NASH was diagnosed with the FLIP algorithm. HCC diagnosis and tumor characteristics, including size and differentiation grade, were obtained from pathology reports.

[0117] All MRIs were taken at a single center (Radiology department, Beaujon University Hospital Paris Nord, Clichy, France). Patients underwent liver examinations on a Philips Ingenia 3T system (Philips, Beaujon, The Netherlands) with gradient amplitude 40 mT / m, a 32-channel phased array body coil, and multi-transmit parallel radiofrequency (RF) transmission technology. A 3D mDIXON sequence with parallel imaging and SENSE (sensitivity encoding) was used for imaging with the following parameters: repetition time (TR) 10.3187 ms, flip angle (FA) 5°, bandwidth 2653 Hz / pixel, echo times: 1.1520, 2.3020, 3.4520, 4.6020, 5.7520, 6.9020, 8.0520, and 9.2020 ms. The imaging matrix size ranged from 176 × 176 × 43 to 400 × 400 × 61, the lateral in-plane spatial resolution ranged from 1.14 mm to 2.05 mm, and the slice thickness was 4 mm. The imaging time was 20 seconds. PDFF values ​​were T2 * Using the IDEAL method, T2 * Corrections were made for relaxation and magnetic susceptibility effects.

[0118] Regions of interest (ROIs) were manually drawn bilaterally at the third lumbar level in the erector spinae (ES), quadratus lumborum, psoas, oblique, and rectus muscles, and the mean proton density fat fraction (PDFF), reflecting the lipid concentration (%) within the ROI for each muscle, was reported (Figure 7). These measurements were performed by a MN (blinded to liver histology or HCC status) who was experienced in muscle fat infiltration measurements and followed guidelines reported in the orthopedic literature. Two types of ROIs were used: (1) a circular ROI (left in Figure 7), hereafter referred to as a "virtual muscle biopsy", by carefully placing a circular ROI in the center of the muscle area while avoiding the margins, intermuscular adipose tissue, and perivenbred regions, and (2) a "complete" ROI (right in Figure 7), outlined by the anatomical boundaries of the muscle. Of note, due to anatomy, in the erector spinae muscles, the "Complete" ROI includes the iliocostalgia and longissimus muscles as well as the multifidus, whereas the virtual biopsy excludes the multifidus and is therefore specific to the major muscle bundles of the erector spinae muscles. Nevertheless, in both cases, for convenience, the measurements are referred to as "ES". To assess the interobserver reliability of the virtual biopsy, MN (see above) and MDB (an experienced radiologist) placed ES ROIs on the MRIs of a randomly selected subset of patients. The consistency of the measurements was excellent (ICC=0.91, 95% CI 0.83-0.95, n=48, p<0.001). Absolute lipid content was correlated with the total muscle area (mm 2 ) was calculated by multiplying it by the PDFF within the ROI of the same cross section.

[0119] At the same level as used for muscle analysis, visceral fat area was measured by MN by drawing ROIs in zones lining the muscle bands within the abdomen with a PDFF threshold of ≥ 50%. Histogram-based radiomics features (first-order radiomics features, i.e., mean, variance, skewness, kurtosis, energy, and entropy) were extracted from the “Complete” muscle ROIs by MATLAB algorithms.

[0120] All data are expressed as mean ± standard deviation unless otherwise stated. Statistical analysis was performed using two-tailed Student's t-test (equal variances) or Welch's t-test (unequal variances), one-way analysis of variance followed by Benjamini Hochberg's post-hoc, chi-square test, and receiver operating characteristic (ROC) curve analysis using Graph Pad Prism 8 software (San Diego, USA). The optimal cut-off of the ROC was determined by the Youden index and the corresponding sensitivity and specificity, odds ratio, and relative risk calculated by the Wilson / Brown, Baptista-Pike, and Koopman asymptotic score methods, respectively. Multivariate analysis was performed with SPSS v24 (IBM, New York, USA) by multiple linear and binary logistic regression. Differences were considered significant at a value of p < 0.05.

[0121] result The total number of patients was 72, with a mean age of 57 ± 14 years and a higher proportion of men (65%, 47 / 72) (Figure 9). NASH, determined histologically by the FLIP algorithm on liver biopsy or resection specimens, was found in 53% of patients. Forty-five patients had focal disease (HCC 20 / 45; 44%). HCC and cirrhosis (F4) were found mainly in NASH patients (Figure 9). However, there was no difference in the prevalence of HCC according to the severity of fibrosis (Figure 10).

[0122] Patients were stratified according to the presence of focal lesions and their subtypes, and muscle-fat infiltration was evaluated by MRI-PDFF-based virtual muscle biopsy. When focal lesions were diagnosed as HCC, patients had significantly higher PDFF in the back muscles (i.e., quadratus lumborum and ES), especially in the ES, compared with patients with another type of tumor (i.e., benign, cholangiocarcinoma, or metastasis) or without focal lesions (Figure 11). Muscle PDFF was similar between patients with benign tumors, cholangiocarcinoma, or liver metastases. Therefore, the analysis was resumed considering two groups of patients: HCC patients and non-HCC patients. HCC patients were older (53±14 vs. 67±9, p<0.001) and had a higher visceral fat area (202±109cm) than non-HCC patients. 2266±117cm 2 ) (Figure 4). HCC patients had significantly higher MRI-PDFF in the back muscles compared to non-HCC patients (4.7±3.6% vs. 3.0±2.2% and 9.6±5.5% vs. 5.7±3.0% in quadratus lumborum and ES, respectively, p<0.02). PDFF in other abdominal muscle bands (i.e., psoas, obliques, and rectus) did not differ between patients with and without HCC.

[0123] To test whether NAFLD status influenced the relationship between muscle fat infiltration and HCC, patients were stratified according to whether they were in the NASH or non-NAFL group (see Methods section). In the NAFL group, muscle PDFF was similar in patients with and without HCC (Fig. 14). In contrast, muscle fat infiltration was two-fold higher in patients with NASH and HCC compared to patients with NASH but without HCC (11 ± 5% vs. 5.4 ± 3.1% in ES) (Fig. 8a-b). ES presented the largest difference (Δ103%, p<0.001) in PDFF (expressed as ESPDFF) in NASH patients with HCC versus those without HCC and was used as the reference muscle for further analysis. Of note, ES muscle area was comparable between patients with NASH and HCC and those with NASH but without HCC (Fig. 15a). This finding is consistent with the higher total lipid content (and therefore lower lean muscle mass) in patients with NASH and HCC compared to patients with NASH but without HCC (Figure 15b).

[0124] In addition to HCC, multiple possible factors may determine myofibril infiltration, such as age, sex, visceral fat area, hepatic inflammatory activity, and fibrosis stage. Therefore, a linear multiple regression model was applied to define predictors of ESPDFF in the entire cohort (n = 72). Significant determinants of ESPDFF in NAFLD patients were age, sex, and HCC, but not activity score, fibrosis score, or visceral fat area.

[0125] Multivariate analysis was used to test whether ESPDFF could significantly predict HCC, independent of possible confounding factors. Adjusted for age, sex, visceral fat area, liver disease activity (histological scores of inflammation and ballooning), and fibrosis stage, ESPDFF remained an independent significant predictor of HCC in NAFLD patients, regardless of the model considered (Figure 6). The association of ESPDFF with HCC was stronger when considering only NASH patients. ROC analysis showed that ESPDFF, whether alone or combined with other parameters, predicted HCC in NASH patients with high performance (area under the ROC curve (AUC): 0.79-0.94, all p<0.05). In NASH patients, ESPDFF ≥ 9% (Youden index) and ≥ 10% (optimal specificity) were associated with a higher relative risk of HCC, of ​​2.7 and 6.4, respectively (Figure 12).

[0126] The size and differentiation status of HCC in patients varied. HCC patients were further stratified by HCC size (≧ or <3 cm) or differentiation grade (well or moderately differentiated). ESPDFF (FIGS. 16a, 16c) and ES total lipid content (FIGS. 16b, 16d) did not differ by HCC size and pathological differentiation grade.

[0127] The mean PDFF measured by virtual liver biopsy was higher in patients with NASH and HCC compared to patients with NASH but without HCC in ES and quadratus lumborum (Fig. 8). Similarly, when PDFF was measured in the whole muscle (i.e., complete ROI), the mean PDFF was higher in patients with NASH and HCC compared to patients with NASH but without HCC in ES and quadratus lumborum (Fig. 9a-9c). We further extracted primary radiomic features from complete ROI (see Methods section). Muscle energy and entropy (Fig. 9a-9c), but not variance, skewness, or kurtosis (not shown), had higher power to distinguish HCC from non-HCC patients. For ES, the AUROCs for energy and entropy were 0.92 (95% CI 0.82-1.00) and 0.88 (95% CI 0.76-0.99, p<0.0001), respectively, to distinguish HCC and non-HCC patients. Of note, in the NASH population, ES energy and ES entropy were independent predictors of HCC in multivariate model 3 of the table in Figure 6 (p=0.047 and p=0.035, excluding ESPDFF). The optimal cutoff for the parameter ES energy (Youden index) to distinguish HCC and non-HCC patients in the NASH population was 0.017, with a sensitivity of 0.96 (95% CI 0.79-0.99) and a specificity of 0.75 (95% CI 0.47-0.91). Furthermore, the combination of at least one clinical factor (e.g., age and sex) and one primary radiomics index of skeletal muscle (e.g., energy in erector spinae) shows an AUROC of 0.97 for energy to distinguish NASH patients with HCC from those without HCC (data not shown).

[0128] Example 2: Materials and Methods All measurements are performed as in Example 1, but on the psoas instead of the erector spinae. The same group of patients as in Example 1 is recruited in this example.

[0129] result NASH patients had higher myo-fatty infiltration heterogeneity compared with NAFL patients (Fig. 17a).

[0130] Primary radiomics features were further extracted from the complete ROI (see Methods section). As shown in Fig. 17b and Fig. 17c, the muscle energy extracted from the psoas muscle has high power to distinguish NASH patients from non-NASH patients. In the psoas muscle, the muscle energy used in the method to distinguish NASH patients from non-NASH patients shows an AUROC of 0.76 (Fig. 17d). The machine learning algorithm described in the present invention (e.g., binary logistic regression) based on the combination of at least one clinical factor (e.g., age and sex) and one primary radiomics index of skeletal muscle (e.g., energy in psoas muscle) shows an increased AUROC of 0.94 to distinguish NASH patients from non-NASH patients (Fig. 17e).

[0131] Example 3: Materials and Methods Patients were prospectively recruited from an outpatient liver clinic. Inclusion criteria were as follows: - Age: 18~75 years old - Presence of steatosis indicated by an elevated FibroScan CAP (controlled attenuation parameter) value (>252 CAP) and / or alanine aminotransferase serum levels >25 and >33 in women and men, respectively.

[0132] Patients meeting the following criteria were excluded from the study: - Liver disease of other etiologies: Hepatitis B surface antigen or HCV RNA positivity, Wilson's disease, hemochromatosis, a1-antitrypsin deficiency, alcoholic liver disease, etc. - Active intravenous drug dependence - Previous diagnosis of cirrhosis - Pregnant women - Medications that can cause fatty liver (e.g. methotrexate, amiodarone, tamoxifen, oral steroids) - Active cancer - Heavy drinking - Renal insufficiency

[0133] Patients were assessed for weight, height, and history of macro- and microvascular complications.Fasting blood samples were collected to measure metabolic parameters (glycemia, insulinemia, triglyceride, and cholesterol profiles), CK-18, and liver disease scores, including fibrosis 4 (FIB-4) and NAFLD fibrosis score (NFS).

[0134] Muscle radiomics indices were assessed non-invasively on MRI-PDFF sequences at the third lumbar level (e.g., to assess the back muscles) and lower extremities (e.g., to assess the leg muscles). Image segmentation and radiomics extraction were performed with the open-source software medical imaging interaction toolkit (MITK).

[0135] Liver tissue was collected for histological diagnosis and for disease staging and grading according to the NAFLD activity score (NAS). Cardiovascular disease risk was calculated with the SCORE2 risk prediction algorithm.

[0136] result We further extracted primary radiomic features from the complete ROI. As shown in Figures 18a, 18b, and 18c, the muscle energy, kurtosis, and entropy extracted from the erector spinae muscle are highly correlated with the NAFLD fibrosis score, which indicates the severity of NAFLD.

[0137] As shown in Fig. 19a, there is a strong correlation between the skewness of muscle radiomics features in leg muscles and the histological NAFLD activity score (NAS). Furthermore, the skewness of radiomics features in lower back muscles (Fig. 19b) and leg muscles (Fig. 19c) as well as the energy of radiomics features (Fig. 19d) are higher in patients with signs of liver inflammation (histological inflammation score 1 or 2) compared to patients without signs of liver inflammation.

[0138] Finally, there is a strong correlation between the variance of muscle radiomics features in the erector spinae muscles and SCORE2, which indicates the risk of cardiovascular disease 10 years later (Figure 20).

[0139] Example 4: Materials and Methods Patients were prospectively recruited from an outpatient liver clinic. Inclusion criteria were as follows: - Age: 18~75 years old - Presence of steatosis indicated by an elevated FibroScan CAP (controlled attenuation parameter) value (>252 CAP) and / or alanine aminotransferase serum levels >25 and >33 in women and men, respectively.

[0140] Patients meeting the following criteria were excluded from the study: - Liver disease of other etiologies: Hepatitis B surface antigen or HCV RNA positivity, Wilson's disease, hemochromatosis, a1-antitrypsin deficiency, alcoholic liver disease, etc. - Active intravenous drug dependence - Previous diagnosis of cirrhosis - Pregnant women - Medications that can cause fatty liver (e.g. methotrexate, amiodarone, tamoxifen, oral steroids) - Active cancer - Heavy drinking - Renal insufficiency

[0141] Patients were assessed for weight, height, and history of macro- and microvascular complications.Fasting blood samples were collected to measure metabolic parameters (glycemia, insulinemia, triglyceride, and cholesterol profiles), CK-18, and liver disease scores, including fibrosis 4 (FIB-4) and NAFLD fibrosis score (NFS).

[0142] Muscle radiomics indices were assessed non-invasively on MRI-PDFF sequences at the third lumbar level (e.g., to assess the back muscles) and lower extremities (e.g., to assess the leg muscles). Image segmentation and radiomics extraction were performed with the open-source software medical imaging interaction toolkit (MITK).

[0143] Liver tissue was collected for histological diagnosis and for disease staging and grading according to the NAFLD activity score (NAS). Cardiovascular disease risk was calculated with the SCORE2 risk prediction algorithm.

[0144] result Various scores have been generated to predict the risk of developing steatohepatitis and / or its complications (ie, HCC and cardiovascular disease).

[0145] The score for predicting the risk of developing steatohepatitis is based on at least one clinical parameter (e.g., age, sex) and at least one radiomics feature (e.g., energy) and at least one other radiomics feature (e.g., variance) in at least one muscle (e.g., erector spinae, leg muscle).

[0146] Risk of having steatohepatitis based on leg muscles (%) = 101.717 (constant) + 0.051 age - 50.569 * Gender(0 / 1)-32.482 * Energy (Z score) +51.792 * Skewness (Z score) + a 1 * x 1 +a 2 * x 2 +a n * x n .

[0147] The score for predicting the risk of developing one complication of steatohepatitis (e.g., HCC or cardiovascular disease) is based on at least one clinical parameter (e.g., age, sex), at least one radiomics feature (e.g., energy) and at least one other radiomics feature (e.g., variance).

[0148] Risk of developing HCC based on erector spinae (%) = -6.661 (constant) + 0.08 * Age +0.923 * Gender(0 / 1)-0.613 * Energy (Z score) +0.031 * Entropy (Z score) + a 1 * x 1 +a 2 * x 2 +a n * x n .

[0149] Risk of developing cardiovascular disease based on erector spinae (%) = -10.282 (constant) + 0.26 * Age +2.307 * Gender (0 / 1)+0.735 * Energy (Z score) +0.094 * Variance(Z score)+a 1 * x 1 +a 2 * x 2 +a n * x n .

[0150] In the above three scores, a n is a constant coefficient, and x n can be (1) a clinical / anthropometric / biological parameter, (2) a first or second order gray level statistic (radiomics) derived from the muscle of interest, or (3) a Z-score normalization of (1) or (2).

Claims

1. A computer-implemented method for determining the risk of developing or having the target non-alcoholic steatohepatitis and / or its complications, comprising: - receiving (11) at least one image (20) of the subject previously acquired by medical imaging technology and at least one clinical and / or biological data (21) previously acquired from the subject; - calculating (12) a fatty infiltration heterogeneity score on at least one image, the fatty infiltration score being calculated based on at least one radiomics feature obtained for a region of interest (ROI) defined on at least one image so as to include at least one skeletal muscle, the at least one radiomics feature being energy, skewness, entropy, kurtosis, or variance; - combining (13) the calculated fatty infiltration heterogeneity score with at least one clinical and / or biological data (21) previously acquired from the subject; - determining (14) the risk (30) of the subject developing or having non-alcoholic steatohepatitis and / or its complications based on the combination of the calculated fatty infiltration heterogeneity score and the at least one clinical and / or biological data (21); wherein the step (13) of combining the calculated fatty infiltration heterogeneity score with at least one clinical and / or biological data previously acquired from the subject includes using a machine learning model.

2. The energy is calculated as follows, 【Number 1】 The entropy is calculated as follows, [Number 2] The skewness is calculated as follows, 【Number 3】 The kurtosis is calculated as follows, 【Number 4】 The variance is calculated as follows 【Number 5】 The method according to claim 1.

3. The method according to claim 2, wherein the fatty infiltration score is calculated based on energy and at least one other radiomics feature.

4. The method according to claim 3, wherein the at least one other radiomics feature is a primary gray level statistic obtained from image pixels or at least one region of the image.

5. The method according to claim 4, wherein the at least one other radiomics feature quantity is selected from mean absolute deviation, root mean square, uniformity, minimum intensity, maximum intensity, mean intensity, median, intensity range, intensity dispersion, intensity standard deviation, skewness, kurtosis, variance, and entropy.

6. The method according to claim 3, wherein the at least one other radiomics feature quantity is a second-order gray level statistic obtained from a co-occurrence matrix of an image, and the second-order gray level statistic is contrast, correlation between adjacent image regions or pixels, energy, first type of uniformity, second type of uniformity, inverse difference moment, Sum average, Sum variance, Sum entropy, autocorrelation, Cluster prominence, Cluster shade, Cluster tendency, dissimilarity, normalized inverse difference moment, normalized inverse difference, inverse variance; Run-length gray level statistics, Short run emphasis, Long run emphasis, Run percentage, Gray-level non-uniformity, Run length non-uniformity, Low gray level run emphasis, High gray level run emphasis, perimeter length, cross-sectional area, major axis length, maximum diameter, and volume and other shape and size-based features; Small area emphasis, Large area emphasis, Intensity variability, Size-zone variability, Zone percentage, Low intensity emphasis, High intensity emphasis, Low intensity small area emphasis, High intensity small area emphasis, Low intensity large area emphasis, and High intensity large area emphasis and other gray-level size-zone matrix-based feature quantities.

7. The method according to claim 3, wherein the fatty infiltration inhomogeneity score is a function of energy and at least one other radiomics feature, each weighted by a weighting factor.

8. The method according to claim 1, wherein the at least one clinical data is selected from age, gender, and personal or family history of metabolic-related adverse events.

9. The method according to claim 1, wherein the machine learning model is a random forest.

10. The method according to claim 1, wherein the complications of non-alcoholic steatohepatitis are selected from hepatocellular carcinoma (HCC), fibrosis (e.g., liver fibrosis), cirrhosis, and cardiovascular disease.

11. The method according to claim 1, wherein the subject has a metabolic disorder, obesity, metabolic syndrome, overweight, and / or fatty liver disease, particularly non-alcoholic fatty liver disease (NAFLD).

12. The method according to claim 1, wherein the subject has non-alcoholic steatohepatitis (NASH).

13. The method according to claim 1, wherein the medical imaging technique is computed tomography, magnetic resonance imaging, or ultrasound.

14. A device (1) for determining the risk of a subject having or developing non-alcoholic steatohepatitis and / or its complications, - at least one input configured to receive at least one image (20) of the subject previously obtained by a medical imaging technique and at least one clinical data and / or biological data (21) previously obtained from the subject, - at least one processor, · calculating (42) a fatty infiltration inhomogeneity score on at least one image, the fatty infiltration score being calculated based on at least one radiomics feature obtained for a region of interest defined on at least one image so as to include at least one skeletal muscle, the at least one radiomics feature being energy, · combining (43) the calculated fatty infiltration inhomogeneity score with at least one clinical data and / or biological data (21) previously obtained from the subject, · determining (44) the risk (30) of the subject having or developing non-alcoholic steatohepatitis and / or its complications based on the combination of the calculated fatty infiltration inhomogeneity score and the at least one clinical data and / or biological data, - at least one processor configured as such. - at least one output adapted to provide said risk (30); A device comprising. **Claim 15**: A computer program for causing a computer to execute the method according to any one of claims 1 to 13.

Images