Method and system for estimating risk of a multiple organ pathology

Abstract

A method and system provide means to assess the likelihood, risk, or possible occurrence of a pathology that impacts multiple organs and / or systems, such as endocrine, exocrine, hypertension, blood and so forth that incorporates organ specific information and its relative role in and on the systemic risk. The method includes collecting medical metrics from the person, the medical metrics comprising quantitative parameters of phenotype of the person, and quantitative measures of a plurality of organs of the person from at least one MRI scan performed to obtain MRI data of the organs, using the metrics to assess a state of each of the organs, and selecting a plurality of the states and at least one of the quantitative parameters to estimate the risk of the multi-organ pathology in the person.

Description

[0001]. METHOD AND SYSTEM FOR ESTIMATING RISK OF A MULTIPLE ORGAN PATHOLOGY Field of the Invention The present invention relates to a method and system for estimating the risk and / or predicting the occurrence of a pathology and / or clinical event. In particular, the present invention provides means to assess the likelihood, risk, or possible occurrence of a pathology that impacts multiple organs and / or systems,referred to as a multi-organ pathology. Organs and systems affected may include, forexample, a blood circulatory system and organs in it, liver, pancreas, kidneys, heart,and so forth. For example, systems may be endocrine or exocrine or respiratory orcirculatory. Hypertension is a multi-organ disease state. The means to assess thelikelihood, risk, or possible occurrence of the multi-pathology incorporates organspecific information and its relative role in and on the systemic risk. In an embodiment the present invention provides a method in which a singlemeasurement or a few specific measurements of the health / status on each one ofseveral individual organs serves as a basis for foresight into a likelihood, risk orprediction of occurrence of a systemic pathology / clinical event, Each of the organsupon which the single measurement or few measurements are made may be part of asystem, such as a blood circulatory system in which all the organs have a role to fulfill.People for whom the invention is applicable may be healthy or feel healthy at the time of the measurement(s) but have a subclinical organ impairment that may be picked upby the measurement(s). The measurement(s) may be from an MRI scan that isperformed on a person to collect data to formulate an image of organs in the system.Background The term ‘pathology’ is used here to mean a condition that may impact multiple organs, i.e. a multi-organ pathology, including where such a ‘pathology’ may also not manifest itself in specific organs but in a systemic way, such as hyperglycaemia, hypertension, metabolic dysfunction etc. Likelihood, risk, or possible occurrence are referred to together as ‘risk’. . Many, and an increasing number of, pathologies impact on multiple organs, for example, Type-2 Diabetes (T2D), cardiovascular disease (CVD), Long Covid, sleep apneoa, and others. Pathologies that impact on multiple organs may be termed multi- organ pathologies. This necessitates the measurement of organ ‘state’ (health, presence and extent of pathology) in each of the organs potentially impacted.An individual person may have several organs affected by the pathology, and todiffering extents.Many methods and systems are known to assess and / or calculate risk of disease in aspecific organ. Methods and systems have been disclosed that combine different measurements to assess a status of a target organ: for example, a combination of measurements of the brain to assess the ‘brain age’ and the susceptibility for disease. Methods are known where organ feature data is used to determine treatment and treatment related risk, including probability of toxicity for a specific organ. Methods have been applied to classify asymptomatic patients into a risk category for a category of disease, for example, the risk of having or developing cancer and / or classifying a patient with an increased risk of having or developing cancer into an organ system-based malignancy class. Methods are known that evaluate a generality: for example, a category of physiology such as an individual’s phenotype / body composition and associated propensity for disease relative to stored population data / profiles. For example, values from MRI scans relating to fat / adipose tissue (the amount of visceral fat, subcutaneous fat, the volume of fat, the concentration of fat in tissue (of an organ and / or bone marrow) have been combined with the phenotype or phenotype variable (such as glycemic status, blood pressure, age or health state) characteristics to determine risk of a future disease. Methods that rely on artificial intelligence (AI) entail the mining and analysis ofextensive data, population, image features and synthetic data for diagnostic,prognostic and predictive information. For example, methods are known to rely onmeta data and aggregate series level data (combinations of patient studies, clinical contextual information, statistical data) as an input to a classifier. Some incorporate . natural language processing. Some methods disclose improved risk by analysing a whole organ from a lesionagnostic perspective. Imaging information may be derived from an MRI scan that isperformed to collect data to formulate an image. The imaging information, clinicalimaging metadata relating to the MRI scan or data collected and additional informationrelating to the pathology or potential tumour type and tumour vicinity may be combined.Many disclosures that claim multi-organ solutions, are confounded by the complexity of segmentation.Certain conditions or multi-organ pathologies, for example metabolic conditions ormulti-organ pathologies, benefit from the simultaneous indication of multiple-organ orsystem state: for example. Assessment of the exocrine, respiratory and circulatorysystems may be via constituent organ status. The state of the organ may be or bedependent upon the organ’s status.However, clinicians can be inundated with myriad data that is at best of very limitedinformative use. Furthermore, where clinicians are specialised, the lack of interpretive systemic relevance can render such data inappropriate and potentially misleading.For example, when observed in isolation neither qualitative measures (e.g. extent ofmass spiculation) or quantitative measures (e.g. tumour dimension) reflect thecomplexity of morphology or behavior within systems of a multi-organ pathology.Where such measures are based on average values for example, average values overa region then a lack of relevance may be aggravated.Products are commercially available that report measures of the states of a number oforgans. Some of the products report measures of a number of quantitativemeasurements (over one hundred) from medical imaging across multiple organs|⋃^ M^ |. The range of organs individually includes the liver, kidneys, pancreas, heart,spleen, aorta and lungs; and collectively extends to measurement of inflammatory conditions such as ‘Long Covid’ which in many patients, impacts several organs including the brain. A number of “relevant parameters” may also be impacted in theseconditions or pathologies including biomarkers such as enzymes in blood serum(alanine transaminase [ALT], alanine transferase [AST], Bilirubin etc). The relevant parameters may encode epidemiologically relevant information such as age, gender etc. and / or genomic information. .However, clinicians find it almost impossible to use this plethora of information. Nosolution exists where the status and respective role of accurate single organ measuresinforms systemic assessment in a useful, informative manner for the clinician. The present invention offers such a solution. It provides means to calculate the risk of a pathology and / or clinical event as it impacts multiple organs and / or systems, such as endocrine system, exocrine system, circulatory system, hepatobiliary system etc. Summary of the Invention According to a first aspect of the invention there is a method of estimating a risk of a multi-organ pathology in a person, comprising: collecting medical metrics from the person, the medical metrics comprising quantitative parameters of phenotype of the person, and one or more quantitative measures of each one of a plurality of organs of the person from at least one MRI scan performed to obtain MRI data of the organs,using the metrics to assess a state of each one of the organs, and selecting a pluralityof the states and at least one of the quantitative parameters to estimate the risk of the multi-organ pathology in the person.The method may be used for estimating the risk and / or predicting the occurrence ofa pathology and / or clinical event. A multi-organ pathology risk estimation systemexecutes the method of estimating a risk of a multi-organ pathology in a person. Thesystem comprises a computer, a data receiving device by which the computer receivesthe quantitative parameters and the quantitative measures of a plurality of organs of the person from at least one MRI scan, a display by which the computer shows a numerical or graphical visualization of the risk, and a computer program directly loadable into a memory of the computer for controlling the method when the program is run on the computer. An MRI scanner may be configured to acquire MRI scans of a plurality of organs in a person, wherein the MRI scanner is in electronic communication with a multi-organ pathology risk estimation system. The method and system provide means to assess the likelihood, risk, or possibleoccurrence of a condition or pathology that impacts multiple organs and / or systems,such as endocrine, exocrine, hepatobiliary, circulatory etc, that incorporates organ specific information and its relative role in and on the systemic risk. In an embodiment the present invention provides a method to transform multiple specific measurements of the health / status of an individual organ or multiple organs . affected by a multi-organ pathology into a likelihood, risk or prediction of occurrence of a systemic pathology / clinical event. The measurements may be quantitative measurements from MRI medical imaging. The method of estimating the risk may include selecting the states of all the organs.The method may use all of the metrics to assess every state. Preferably the methoduses a selection of the metrics that the multi-organ pathology directly affects to assessthe state of each of the organs. The metrics may comprise a selection of the quantitative measures from each of the organs that the multi-organ directly affects. However, preferably the method includes selecting the states particularly from other than those of the organs that the multi-organpathology directly affects plus those that the multi-organ pathology directly affects. Thelist of organs may not be obvious from traditional clinical teachings ~ e.g. adding in acombination of liver and pancreas state or liver, kidneys, and pancreas state to improveprediction of cardiac outcomes – something that is not part of standard clinical thought.Wherein the multi-organ pathology is cardiovascular or cardiac, organs other thanthose that the multi-organ pathology directly affects include the kidneys, or the liverand the pancreas plus those the multi-organ pathology directly affects include theheart. Here ‘or’ may be inclusive as in ‘and’.The method may include selecting the states of the organs that the multi-organ pathology directly affects according to levels of the quantitative parameters or quantitative measures of blood, liver, and biliary tree required for autoimmune liver disease.Subsets of measurements may be defined, and their importance may be adjusted, ormay be weighted relative to each other. The subsets may define risk scores tocharacterise the health of specific systems (circulatory, respiratory, exocrine, endocrine, etc.) or clinically relevant events ~e.g. hospitalisation, hepaticdecompensation, etc. Therefore, the method may assess the state of each of theorgans using a weighted sum of the metrics. The method may include assessing therespective state of each of the organs using a weighted sum of a plurality of thequantitative measures. The weights may be determined by linear or logisticregression.A logistic regression model may have a dependent variable dependent upon . independent variables, wherein the dependent variable is for status of the multi-organ pathology and the independent variables are respective to each one of the medical metrics. The method may include reducing the logistic regression model by removing from the logistic regression model at least one of the independent variables by a stepwise backwards selection process removing the independent variables one at a time sequentially to identify a parsimonious model based on Akaike Information Criterion optimization.The method may use a point in a space having a number of dimensions equal to howmany there of the quantitative measures of the each of the organs, wherein an n-cube that represents the states that are normal healthy is calculated, and clustering is performed within the space to quantify the sensitivity of the risk to each of the quantitative measures of each of the individual organs.The method may use a Bayesian network in which the metrics for each of the organsare nodes, assigning a categorical value to each of the nodes, and propagating the categorical values through the Bayesian network to estimate the risk. At the time of the measurement(s), the person may have only subclinical organ impairment. In the future, the multi-organ pathology may progress in the person uponwhom the measurement(s) are taken. In the future the person may require treatmentfor the multi-organ pathology or organs which currently have subclinical organ impairment. The method may provide a quantitative indication of risk of future occurrence of the multi-organ pathology developing in the person that may require treatment of the multi- organ pathology.The MRI data may be standardized or adjusted to that of a reference MRI scanner.For example, the data may be standardized to a reference main magnetic field level.Thus the measurement(s) or quantification of the risk may also be standardized.The method may turn all of the metrics except age into thresholded, categoricalvariables using clinically defined cutoffs. The method may include monitoring separate groups of healthy people in a population.Those of the healthy people who go on to be diagnosed with disease or the multi-organ .pathology are recorded and the group that they are in is recorded. Each group maybe characterized by a specific combination of the metrics. This information may beused in the method of quantifying the risk of the disease or multi-organ technologybeing diagnosed or occurring according to specific combinations of the metrics or levels of the metrics in each combination. The information from the population and separate groups of healthy people within the population may quantify a risk that different subsets of healthy people will become unhealthy according to the metrics or levels of the metrics in each combination which characterize the group. It may be used to as guide to quantify the risk that the person upon whom the measurements are made and for who the metrics are determined will develop the multi-organ pathology or disease in the future.The method of estimating the risk may include identifying a relevant population pool ofpeople who are initially healthy and monitoring over time the same medical metrics and the state of each of the same organs of each one of the people of the relevantpopulation pool as are collected from the person; recording quantitative levels of themetrics and the state of each of the organs of the people in the relevant populationpool over time; and correlating a rate of change in each of the metrics and the state ofeach of the organs of the people in the relevant population pool with a rate of occurrence of the multi-organ pathology in the people and using the correlation to adjust the estimate of the risk of the multi-organ pathology in the person. The risk estimate is thereby made robust because the pool of people accounts for variations in individual susceptibility to the multi-organ pathology.The method of estimating the risk may include identifying a relevant population pool ofpeople in which an indication that the multi-organ pathology is present develops overtime from an initial indication that the multi-organ pathology is not present, and identifying a control population pool without the indication that the multi-organpathology is present, collecting the same medical metrics and assessing the samestates for the people in the relevant population pool and for people in the control population as are collected from the person, and determining a difference between the states in the people in the relevant population compared to the control population, correlating each difference with a rate of occurrence of the multi-organ pathology inthe relevant population pool over the time, and using the correlation to adjust theestimate of the risk of the multi-organ pathology in the person. The risk estimate is .thereby made dependable because the control population provides a statisticalbaseline. According to an aspect of the invention there is a method of estimating the risk of a multi-organ pathology in a person wherein the risk is developing diabetic hyperglycaemia, comprising: collecting medical metrics from the person, the medical metrics comprising quantitative parameters of phenotype of the person including bodymass index BMI, gender, and glycerated haemoglobin HbA1c, and one or morequantitative measures of each one of a plurality of organs of the person from at least one MRI scan performed to obtain MRI data of the organs including visceral adipose tissue VAT, subcutaneous adipose tissue SAT, skeletal muscle tissue SMI, liver fat,and liver iron corrected-T1 CT1, using the metrics to assess a state of each of theorgans, and selecting a plurality of the states and at least one of the quantitativeparameters to estimate the risk of the multi-organ pathology in the person, using aBayesian network in which the metrics for each one of the organs are nodes, assigninga categorical value of either low or elevated to each of the nodes, and propagating the categorical values through the Bayesian network to estimate the risk as probability of a future occurrence of hyperglycemia wherein HbA1c exceeds 48 mmol / mol.According to an aspect of the invention there is a method of estimating the risk of amulti-organ pathology in a person wherein the risk is developing a cardiovasculardisease or cardiac disease, comprising: collecting medical metrics from the person,the medical metrics comprising quantitative parameters of phenotype of the person, and one or more quantitative measures of each one of a plurality of organs of theperson from at least one MRI scan performed to obtain MRI data of the organs, thequantitative measures including: iron-corrected T1 data of the liver indicative of fibro- inflammation, PDFF data of the liver indicative of liver fat, scanner-referenced T1 dataof the pancreas indicative of fibro-inflammation, the quantitative parameters including:the person’s age, body mass index BMI, their type or value of blood and urine biochemistry including quantitative values of at least one circulating biomarker:aspartate aminotransferase (AST), alanine aminotransferase (ALT), glyceratedhaemoglobin (HbA1c), or C-reactive protein, using the metrics to assess a state ofeach one of the organs, and selecting a plurality the states and a plurality of the quantitative parameters to estimate the risk of the multi-organ pathology in the person. The method may include identifying a relevant population pool of people who are .initially healthy and monitoring over time the same medical metrics and the state ofeach of the same organs of each one of the people of the relevant population as arecollected for the person, and correlating a rate of change in each of the metrics or thestate of each of the organs in the people in the relevant population pool, or both themetrics and the states, with a rate of occurrence of the cardiovascular disease orcardiac disease in the people in the relevant population pool, and using the correlationto adjust the estimate of the risk of the multi-organ pathology in the person. Separategroups of initially healthy people in the relevant population may be compared andtracked over time. Each of the groups may comprise people who have metrics or states of each of their organs that are different from those of people in the other groups. By comparing the groups, their metrics, and their states, how different combinations ofthe metrics and states quantify risk to the multi-organ pathology, for examplecardiovascular disease or cardiac disease, may be determined.The method may identify a relevant population pool of people in which an indicationthat the cardiovascular disease or cardiac disease is present develops over time froman initial indication that the cardiovascular disease or cardiac disease is not present.The method may identify a control population pool without the indication that thecardiovascular disease or cardiac disease is present. The method may collect thesame medical metrics and assess the same states for the people in the relevantpopulation pool and for people in the control population as are collected from theperson, and determine a difference between the states in the people in the relevantpopulation compared to the control population. The method may include correlatingeach difference with a rate of occurrence of the cardiovascular disease or cardiacdisease in the population pool over the time that the cardiovascular disease or cardiacdisease develops, and using the correlation to adjust the estimate of the risk of themulti-organ pathology in the person. According to an aspect of the invention there is a method of estimating the risk of a multi-organ pathology in a person wherein the risk is of developing diabetic macular oedema DMO, comprising: collecting medical metrics from the person, the medical metrics comprising quantitative parameters of phenotype of the person, and one or more quantitative measures of each one of a plurality of organs of the person from atleast one MRI scan performed to obtain MRI data of the organs, the quantitativemeasures including: iron-corrected T1 data of the liver indicative of fibro-inflammation, PDFF data of the liver indicative of liver fat, scanner-referenced T1 data of the . pancreas indicative of fibro-inflammation, the quantitative parameters including a demographic, a feature of the person’s medical history relevant to DMO, their type or value of blood and urine biochemistry, using the metrics to assess a state of each one of the organs, and selecting a plurality the states and a plurality of the quantitative parameters to estimate the risk of the multi-organ pathology in the person.The invention will now be described, by way of example only, with reference to theaccompanying figures in which: Brief Description of the Figures Figure 1 shows a flowchart summarizing the main steps of the invention; Figure 2 shows a schematic representation of combining the multiple quantitative measurements with relevant parameters to predict single organ and multi-organ pathologies; Figure 3 shows estimation of diabetic hyperglycaemia using a Bayesian network; and Figure 4 shows multi-organ risk of cardiovascular outcomes. Detailed Description of the InventionA flowchart is shown in Figure 1 of method (100) of estimating a risk of a multi-organpathology in a person , comprising: collecting medical metrics from the person, themedical metrics comprising quantitative parameters (104) of phenotype of the person,and quantitative measures (102) of a plurality of organs of the person from at least oneMRI scan performed to obtain MRI data of the organs, using the metrics to assess astate (106) of each of the organs, and selecting a plurality of the states (108) and atleast one of the quantitative parameters to estimate the risk of the multi-organpathology (110) in the person.In an embodiment a number of quantitative measurements mij are made from medicalimaging data, there is a set M (220) of the quantitative measurements:M^ = ^m^^|j = 1, … , J^^of each of a set of organs (210). . {O^|i = 1, … , I}as well as a number of ‘relevant parameters’ (240)C = {c^: l = 1, … , L}.As shown in Figure 2 there are quantitative measurements (221, 222, 223, 224) ofeach organ (211, 212, 213, 214). The quantitative measurements of one of theorgans may be different than the quantitative measurements on another of the organs. Some organs may have quantitative measurements in common with other organs. Figure 2 also shows that there are additional relevant parameters (241, 242, 243, 244) which may not be organ specific.The “relevant parameters” (240) C may be biomarkers such as the enzymes (ALT(327), AST (328), Bilirubin, ….) or they may encode epidemiologically relevantinformation such as age (332), gender (329) etc . They may also encode genomicinformation of the person in the medical imaging scans.Each measurement or parameter may be continuous, for example glyceratedhaemoglobin (HbA1c) blood glucose level (336) or cT1 (331), or it may be categorical(e.g. very low, low, normal, high, very high). It is often the case that clinicians use categorical measures. An error bar may be associated with each measurement, which may or may not bemade explicit (e.g. an iron corrected T1 determined from data collected from an MRIscan of a liver, cT1 = 850ms + / - 20ms).A continuous measure typically has an associated distribution. Such a distribution may be summarised by parameters (e.g. mean and standard deviation in the case of a normally distributed parameter, or mode and left and right spread in the case of a non-normally distributed, asymmetric measure).The quantitative measurement (e.g. cT1 = 900 ms) may either have aprobability / likelihood estimate relative to the distribution or will be assigned acategorical value. For example, parameter, body mass index (BMI) = height (m)divided by weight2(kg2) is continuously varying but is often categorised as below normal, normal, overweight, obese, and morbidly obese, where, for example obesemeans 30 ≤ ^^^ < 40. . An example of such a categorical distribution, which gives the probability of diabetichyperglycaemia (DHG) given the BMI status (335) is as follows : An estimation of diabetic hyperglycaemia (DHG) using a Bayesian network (300) isshown in Figure 3. A similar set of categorical distributions may be established for the medical imagingmetrics, C and The categories can be created by comparison to a normal range(e.g. normal [<upper limit of normal {ULN}], slightly elevated [1-2x ULN], significantly elevated [ >2xULN]) or by previously established empirical thresholds (e.g. cT1<800 is “Normal”, 800 ≤ cT1 ≤ 875 is “nonalcoholic steatohepatitis (NASH)”, cT1 > 875 ms is“high-risk NASH”). The individual measurements, ^^^, are made of each individualorgan ^^.. In an embodiment, the state of each organ O^(211, 212, 213, 214) is determined viathe totality of information that are the quantitative measurements (221, 222, 223, 224)of the respective organ together with the relevant parameters (241, 242, 243, 244) ofthe person, Ω^ = M^ ∪ Cfrom which the state of the organ is determined and a set of pathologies (250, 252,252, 253) specific to a respective organ O^. The multi-organ pathology affects the stateof a plurality of the organs, and the state (231, 232, 233,234) of each of the organ canaffect or indicate the progress or risk of the multi-organ pathology (260, 261, 262, 263). . For example, with reference to the liver, a set of pathologies refer specifically to organ O^include: steatosis, metabolic-associated steatohepatitis, cirrhosis, hepatocellular carcinoma, metastatic liver cancer, jaundice etc. For most of these pathologies no additional information is required about the organs other than O^.The pathologies pik that impact only organ O^ are denoted by: and the risk of one of these pathologies by: p(p^^) = f^^(Ω^),Where not all of the parameters Ω^ are needed to determine p(p^^), a subset ofparameters is identified by:Ω^^ ⊆ Ω^so that p^^ = f^^(Ω^) = f^^(Ω^^).We refer to this herein as ‘pathology projection’ as it defines the reduction in theparameters and quantitative measurements needed to determine the risk to a predetermined level of accuracy. Ω^ → Ω^^The attention of a clinician is drawn to one or more measurements m^^thus guiding the clinician’s focus to the measurements that appear of concern.The state of a specific organ within the set and / or subset Ω^ is required, expressed by:σ^^^Ω^^ and p^^ = f^^(σ^) = f^^(σ^^).For single-organ pathologies clinicians know which subset of the totality of parametersthat are reported for an organ are relevant for a specific pathology p^^. Examples of ‘single-organ’ pathologies include: steatohepatitis, fatty pancreas disease, ischemic- . type biliary lesions (post-liver transplantation), pancreatitis.In an embodiment, the function ^^^(σ^^) from which the risk of the pathology ^^^ iscalculated, comprises any machine learning (ML) technique: for example, where a sufficient number of cases of the pathology, and of “normal” cases (i.e. non-pathology) are available.In an embodiment, the function ^^^(σ^^) from which the risk of the pathology ^^^ iscalculated, may be a weighted sum of the parameters C and quantitative measurementMi in Ω^^. the weights may be learned by an appropriate ML technique.In an embodiment, the function ^^^(σ^^) from which the risk of the pathology ^^^ iscalculated, may be a weighted sum where the weights are not learned, but optimised in a conventional way (e.g. linear or logistic regression).In an embodiment, the ‘function’, ^^^, describes a point in ^ dimensional space, where^ = ^^^^^, or ^^^^^ + |^|. . A distance (e.g. Euclidian, Gower) from an n-cube thatrepresents ‘normal’ or ‘healthy’ states may then be calculated either in a specificdimension, ^^ , or the totality of all dimensions, Additionally, clustering (e.g. k-means, portioning around medoids, …) can be performed within the n-dimensional space to characterise the pathology and highlight the key parameters C or quantitative measurements mij (i.e. those which have the most impact on likelihood of the pathology being present) to a clinician. In a preferred embodiment, the ‘function’ ^^^may be best represented by a Bayesian network (300) in which the parameters and quantitative measurements in Ω^^are thenodes (311, 312, 321, 322, 323, 324, 325, 326, 327, 328, 329, 330, 331, 332, 333,334, 335, 337), each one of which has an associated probability or categorical value and these values / probabilities are propagated through the Bayesian network to estimate the risk / likelihood / possibility of pathology ^^^. In an embodiment, a multi-organ pathology ^^^impacts more than one organ, but not necessarily all of the organs : ^^ ⊆ {^^|^ = 1, … , ^}where the uthsuch pathology, ^^^, is associated with the set ^^of organs. A function . assesses the risk of the pathology from the states of the organs that are affected: where ^^^is the state of the jthorgan relevant to this pathology. In an embodiment, the list of organs may be derived from expert input (e.g. measurements of blood, liver, and biliary tree required for autoimmune liver disease). In an embodiment, the list of organs is derived from data. In an embodiment, the list of organs is not obvious from traditional clinical teachings(e.g. adding in liver state to improve prediction of cardiac outcomes – something notpart of standard clinical thought). The mechanisms to calculate a single organ apply.In an embodiment, a risk of a multi-organ pathology p( ^^^) is calculated using anymachine learning (ML) technique: for example, where a sufficient number of cases of the pathology, and of ‘normal’ cases (i.e. non-pathology) are available.In an embodiment, a risk of a multi-organ pathology p( ^^^ ) is calculated using aweighted sum of the parameters C or measurement Mi in Ω^^and the weights may belearned by an appropriate ML technique.In an embodiment, a risk of a multi-organ pathology p( ^^^ ) is calculated using aweighted sum where the weights are not learned but optimised in a conventional way(e.g. linear or logistic regression).In an embodiment, a risk of a multi-organ pathology p( ^^^) is calculated using a pointin ^ dimensional space, where ^ = A distance (e.g. Euclidian, Gower) from ann-cube that represents ‘normal’ or ‘healthy’ states may then be calculated either in aspecific dimension, ^^ , or the totality of all dimensions, Additionally, clustering(e.g. k-means, portioning around medoids, …) can be performed within the n- dimensional space to characterise the pathology and highlight the key measurements (i.e. those which have the most impact on likelihood of the pathology being present) to a clinician.In a preferred embodiment, a risk of a multi-organ pathology p( ^^^) is represented by .a Bayesian network (300) in which the parameters C and measurement Mi in Ω^^ arethe nodes, each one of which has an associated probability or categorical value and these values / probabilities are propagated through the Bayesian network to estimate the risk / likelihood / possibility of pathology ^^^. With reference to Figure 3, diabetic hyperglycaemia is estimated using a simplifiedBayesian network (300) whereby MR-derived measures Mi of visceral adipose tissue(VAT) (333), subcutaneous adipose tissue (SAT) (330), skeletal muscle index (SMI)(334), liver fat (337) and liver iron corrected-T1 (cT1) (311) are combined with routinelycollected parameters C, for example biomarkers such as body mass index (BMI) (335),sex (329) and glycerated haemoglobin (HbA1c) (336). The risk or probability ofhyperglycemia (HbA1c >48mmol / mol) is estimated after fixing the values of network variables to ‘low’ or ‘elevated’.By way of illustrative example as shown in Figure 4, imaging metrics includingparameters and quantitative measurements are derived from abdominal multi-parametric MRI scans. To assess the level of risk of cardiovascular hospitalisation,quantitative measurements of liver fibro-inflammation (iron-corrected T1, cT1) andliver fat (PDFF) were calculated, alongside quantitative measurements of pancreaticfibro-inflammation (scanner-referenced T1, srT1). All imaging metrics are standardised to be comparable across scanner manufacturers and field strengths. Imaging metrics are combined with parameters including the participants’ age and BMI. Blood biochemistry is used to calculate levels of circulating biomarkers aspartate aminotransferase (AST), alanine aminotransferase (ALT), glycated haemoglobin (HbA1c) and C-reactive protein. Each of the metrics above (with the exception of age) are turned into thresholded variables using clinically defined cutoffs. In addition, each metric is independently associated with an increased risk of cardiovascular disease- related hospitalisation.Thus, a categorical risk model comprises combining metrics with outcome data. Oneexample is a Cox proportional hazard model. This model links changes in metrics to the rate of outcomes. As such, it allows for the calculation of predicted hospitalisation- free survival probability for any combination of input metrics. As a further step, an individual’s predicted 3-year survival probability is used as an easy to interpret risk score. The relative impact of different treatment approaches can be easily analysed by modifying input variables based on its expected effect on modifiable input variables. In . the present example, a drug that normalised HbA1c values alone would have less effect than a drug which was able to reduce fibro-inflammation in the liver and pancreas. By way of illustrative example, the method is used to estimate an individual’s risk ofdeveloping diabetic macular oedema (DMO). This is a complication of diabetes mellitusaffecting the eyes which can lead to vision loss. A number, k, of imaging metrics are collected from an individual undergoing clinical assessment and multi-parametric MRI scan, including quantitative measurements of liver fibro-inflammation (iron-corrected T1, cT1) and pancreatic fibro-inflammation (scanner-referenced T1, srT1), liver fat(PDFF) and parameters covering demographics, medical history, blood and urinebiochemistry. All the imaging metrics are standardised to be comparable across scanner manufacturers and field strengths. A relevant population pool is identified from various data sources, in which the indication of interest (DMO) is present along with controls without the indication. The k metrics from the individual which underwent the clinical assessment are selected from the relevant population pool, to create a truncated dataset. Based on the truncated dataset, a logistic regression model isdefined with DMO status as the dependent variable, y, and all k independent variablesincluded at the beginning to comprise the with the full model. A stepwise backwards variable selection process is utilised, and independent variables are removed one at a time sequentially to identify the most parsimonious model based on Akaike Information Criterion (AIC) optimisation, i.e. the final model. The final model summarising the association between the dependent variable (DMO) and the key independent variables, is used to predict the probability of DMO in the case of the individual’s characteristics. The individual’s probability score can be statistically transformed and presented in a visual manner to contextualise the relative risk of the disease indication and improve patient and clinician understanding. The invention has been described by way of examples only. Therefore, the foregoing is considered as illustrative only of the principles of the invention. Further, sincenumerous modifications and changes will readily occur to those skilled in the art, it isnot desired to limit the invention to the exact construction and operation shown and described, and accordingly, all suitable modifications and equivalents may be resorted to, falling within the scope of the claims.

Claims

. Claims:

1. A method of estimating a risk of a multi-organ pathology in a person, comprising:collecting medical metrics from the person, the medical metrics comprisingquantitative parameters of phenotype of the person, and one or more quantitative measures of each one of a plurality of organs of the person from at least one MRIscan performed to obtain MRI data of the organs, using the metrics to assess a state of each one of the organs, and selecting a pluralityof the states and at least one of the quantitative parameters to estimate the risk ofthe multi-organ pathology in the person.

2. A method of estimating the risk according to claim 1 wherein the states selectedinclude the state of every one the organs.

3. A method of estimating the risk according to claim 1 using all of the metrics to assessevery state.

4. A method of estimating the risk according to claim 1 using a selection of the metricsthat the multi-organ pathology directly affects to assess the state of each of the organs.

5. A method of estimating the risk according to claim 1 wherein the metrics comprise aselection of the quantitative measures of each one of the organs that the multi-organdirectly affects.

6. A method of estimating the risk according to claim 1 wherein the states selectedinclude other than those of the organs that the multi-organ pathology directly affects plus those that the multi-organ pathology directly affects.

7. A method of estimating the risk according to claim 4, 5, or 6 selecting the states ofthe organs that the multi-organ pathology directly affects according to levels of thequantitative parameters or quantitative measures of blood, liver, and biliary tree required for autoimmune liver disease.

8. A method of estimating the risk according to claim 6 wherein the multi-organpathology is cardiovascular or cardiac and the organs other than those that the multi-. organ pathology directly affects include the liver and pancreas, and those the multi-organ pathology directly affects include the heart.

9. A method of estimating the risk according to any preceding claim 1 assessing thestate of each of the organs using a weighted sum of the metrics.

10. A method of estimating the risk according to any preceding claim assessing therespective state of each of the organs using a weighted sum of a plurality of thequantitative measures.

11. A method of estimating the risk according to claim 9 or 10 wherein the weights aredetermined by linear or logistic regression.

12. A method of estimating the risk according to claim 9 or 10 using a logistic regressionmodel having a dependent variable dependent upon independent variables, wherein the dependent variable is for status of the multi-organ pathology and the independentvariables are respective to each one of the medical metrics.

13. A method of estimating risk according to claim 12 reducing the logistic regressionmodel by removing from the logistic regression model at least one of the independent variables by a stepwise backwards selection process removing the independent variables one at a time sequentially to identify a parsimonious model based on Akaike Information Criterion optimization.

14. A method of estimating the risk according to any preceding claim using a point in aspace having a number of dimensions equal to how many there of the quantitative measures of the each of the organs, wherein an n-cube that represents the statesthat are normal healthy is calculated, and clustering is performed within the space toquantify the sensitivity of the risk to each of the quantitative measures of each of the individual organs.

15. A method of estimating the risk according to any preceding claim using a Bayesiannetwork in which the metrics for each of the organs are nodes, assigning acategorical value to each of the nodes, and propagating the categorical values through the Bayesian network to estimate the risk.

16. A method of estimating the risk according to any preceding claim wherein the risk isof future occurrence of the multi-organ pathology developing in the person to require. treatment of the multi-organ pathology.

17. A method of estimating the risk according to any preceding claim standardizing theMRI data to a reference MRI scanner.

18. A method of estimating the risk according to any preceding claim turning all of themetrics except age into thresholded categorical variables using clinically definedcutoffs.

19. A method of estimating the risk according to any preceding claim includingidentifying a relevant population pool of people who are initially healthy and monitoring over time the same medical metrics and the state of each of the same organs of each one of the people of the relevant population pool as are collected from the person, recording quantitative levels of the metrics and the state of each of the organs of the of the people in the relevant population pool over time, and correlating a rate of change in each of the metrics and the state of each of the organsof the people in the relevant population pool with a rate of occurrence of the multi-organ pathology in the people and using the correlation to adjust the estimate of therisk of the multi-organ pathology in the person.

20. A method of estimating the risk according to claims 1 to 18 includingidentifying a relevant population pool of people in which an indication that the multi-organ pathology is present develops over time from an initial indication that the multi-organ pathology is not present, and identifying a control population pool without the indication that the multi-organ pathology is present,collecting the same medical metrics and assessing the same states for the people inthe relevant population pool and for people in the control population as are collected from the person, and and determining a difference between the states in the people in the relevantpopulation compared to the control population, correlating each difference with a rate. of occurrence of the multi-organ pathology in the relevant population pool over thetime, and using the correlation to adjust the estimate of the risk of the multi-organ pathology in the person.

21. A method of estimating the risk of a multi-organ pathology in a person wherein therisk is developing diabetic hyperglycaemia, comprising: collecting medical metrics from the person, the medical metrics comprising quantitative parameters of phenotype of the person including body mass index BMI,gender, and glycerated haemoglobin HbA1c, and one or more quantitative measures of each one of a plurality of organs of the person from at least one MRI scan performed to obtain MRI data of the organs including visceral adipose tissue VAT,subcutinous adipose tissue SAT, skeletal muscle tissue SMI, liver fat, and liver ironcorrected-T1 CT1, using the metrics to assess a state of each of the organs, and selecting a plurality of the states and at least one of the quantitative parameters to estimate the risk of themulti-organ pathology in the person,using a Bayesian network in which the metrics for each one of the organs are nodes,assigning a categorical value of either low or elevated to each of the nodes, and propagating the categorical values through the Bayesian network to estimate the risk as probability of a future occurrence of hyperglycemia wherein HbA1c exceeds 48 mmol / mol.

22. A method of estimating the risk of a multi-organ pathology in a person wherein therisk is developing a cardiovascular disease or cardiac disease, comprising:collecting medical metrics from the person, the medical metrics comprisingquantitative parameters of phenotype of the person, and one or more quantitative measures of each one of a plurality of organs of the person from at least one MRI scan performed to obtain MRI data of the organs,the quantitative measures including: iron-corrected T1 data of the liver indicative offibro-inflammation, PDFF data of the liver indicative of liver fat, scanner-referenced T1 data of the pancreas indicative of fibro-inflammation, the quantitative parameters including: the person’s age, body mass index BMI, their. type or value of blood and urine biochemistry including quantitative values of at leastone circulating biomarker: aspartate aminotransferase (AST), alanineaminotransferase (ALT), glycerated haemoglobin (HbA1c), or C-reactive protein, using the metrics to assess a state of each one of the organs, and selecting a pluralityof the states and a plurality of the quantitative parameters to estimate the risk of themulti-organ pathology in the person.

23. A method of estimating the risk according to claim 22 includingidentifying a relevant population pool of people who are initially healthy andmonitoring over time the same medical metrics and the state of each of the same organs of each one of the people of the relevant population pool as are collected forthe person, recording quantitative levels of the metrics and the state of each of the organs of the of the people in the relevant population pool over time, and correlating a rate of change in each of the metrics and the state of each of the organsin the people in the relevant population pool with a rate of occurrence of thecardiovascular disease or cardiac disease in the people in the relevant populationpool and using the correlation to adjust the estimate of the risk of the multi-organpathology in the person.

24. A method of estimating the risk according to claim 22 includingidentifying a relevant population pool having an indication that the cardiovascular disease or cardiac disease is present,identifying a control population pool without the indication that the cardiovascular disease or cardiac disease is present,collecting the same medical metrics and assessing the same states for people in the relevant population pool and for people in the control population as are collected from the person, and determining a difference between the states in the people in the relevant populationcompared to the control population, correlating each difference with a rate of. occurrence of the cardiovascular disease or cardiac disease in the population pool,using the correlation to adjust the estimate of the risk of the multi-organ pathology inthe person.

25. A method of estimating the risk of a multi-organ pathology in a person wherein therisk is of developing diabetic macular oedema DMO, comprising:collecting medical metrics from the person, the medical metrics comprisingquantitative parameters of phenotype of the person, and one or more quantitative measures of each one of a plurality of organs of the person from at least one MRIscan performed to obtain MRI data of the organs, the quantitative measures including: iron-corrected T1 data of the liver indicative of fibro-inflammation, PDFF data of the liver indicative of liver fat, scanner-referencedT1 data of the pancreas indicative of fibro-inflammation, the quantitative parameters including a demographic, a feature of the person’smedical history relevant to DMO, their type or value of blood and urine biochemistry,using the metrics to assess a state of each one of the organs, and selecting a pluralityof the states and a plurality of the quantitative parameters to estimate the risk of themulti-organ pathology in the person.

26. A multi-organ pathology risk estimation system to execute the method of estimatinga risk of a multi-organ pathology in a person according to any of claims 1 to 25comprising a computer, a data receiving device by which the computer receives the quantitative parameters and the quantitative measures of a plurality of organs of the person from at least one MRI scan, a display by which the computer shows a numerical or graphical visualization of the risk, and a computer program directly loadable into a memory of the computer for controlling the method when the program is run on the computer.

27. An MRI scanner configured to acquire MRI scans of a plurality of organs in a person,wherein the MRI scanner is in electronic communication with a multi-organ pathology risk estimation system according to claim 26.

Images