WO2023031355A1 - Health state estimation using machine learning - Google Patents
Health state estimation using machine learning Download PDFInfo
- Publication number
- WO2023031355A1 WO2023031355A1 PCT/EP2022/074380 EP2022074380W WO2023031355A1 WO 2023031355 A1 WO2023031355 A1 WO 2023031355A1 EP 2022074380 W EP2022074380 W EP 2022074380W WO 2023031355 A1 WO2023031355 A1 WO 2023031355A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- subject
- measurement
- machine learning
- tissue
- health state
- Prior art date
Links
- 238000010801 machine learning Methods 0.000 title claims abstract description 91
- 230000036541 health Effects 0.000 title claims abstract description 44
- 238000005259 measurement Methods 0.000 claims abstract description 111
- 238000000034 method Methods 0.000 claims abstract description 108
- 238000001228 spectrum Methods 0.000 claims abstract description 58
- 230000008569 process Effects 0.000 claims abstract description 56
- 238000012549 training Methods 0.000 claims description 66
- 239000000126 substance Substances 0.000 claims description 28
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 claims description 20
- 239000008103 glucose Substances 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 18
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 11
- HVYWMOMLDIMFJA-DPAQBDIFSA-N cholesterol Chemical compound C1C=C2C[C@@H](O)CC[C@]2(C)[C@@H]2[C@@H]1[C@@H]1CC[C@H]([C@H](C)CCCC(C)C)[C@@]1(C)CC2 HVYWMOMLDIMFJA-DPAQBDIFSA-N 0.000 claims description 10
- 239000000470 constituent Substances 0.000 claims description 10
- JYGXADMDTFJGBT-VWUMJDOOSA-N hydrocortisone Chemical compound O=C1CC[C@]2(C)[C@H]3[C@@H](O)C[C@](C)([C@@](CC4)(O)C(=O)CO)[C@@H]4[C@@H]3CCC2=C1 JYGXADMDTFJGBT-VWUMJDOOSA-N 0.000 claims description 10
- 238000003066 decision tree Methods 0.000 claims description 7
- 230000009467 reduction Effects 0.000 claims description 6
- JVTAAEKCZFNVCJ-UHFFFAOYSA-M Lactate Chemical compound CC(O)C([O-])=O JVTAAEKCZFNVCJ-UHFFFAOYSA-M 0.000 claims description 5
- WQZGKKKJIJFFOK-VFUOTHLCSA-N beta-D-glucose Chemical compound OC[C@H]1O[C@@H](O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-VFUOTHLCSA-N 0.000 claims description 5
- 235000012000 cholesterol Nutrition 0.000 claims description 5
- 229960000890 hydrocortisone Drugs 0.000 claims description 5
- 238000012706 support-vector machine Methods 0.000 claims description 5
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 4
- 230000000306 recurrent effect Effects 0.000 claims description 3
- 210000001519 tissue Anatomy 0.000 description 43
- 239000008280 blood Substances 0.000 description 26
- 210000004369 blood Anatomy 0.000 description 26
- 230000006870 function Effects 0.000 description 22
- 239000011159 matrix material Substances 0.000 description 11
- 206010012601 diabetes mellitus Diseases 0.000 description 9
- 239000000523 sample Substances 0.000 description 9
- 208000010125 myocardial infarction Diseases 0.000 description 7
- 238000012544 monitoring process Methods 0.000 description 6
- 239000013598 vector Substances 0.000 description 5
- 230000000747 cardiac effect Effects 0.000 description 3
- 206010033675 panniculitis Diseases 0.000 description 3
- 210000004304 subcutaneous tissue Anatomy 0.000 description 3
- 241001465754 Metazoa Species 0.000 description 2
- 206010040047 Sepsis Diseases 0.000 description 2
- 208000006011 Stroke Diseases 0.000 description 2
- 238000000862 absorption spectrum Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000036772 blood pressure Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 229940079593 drug Drugs 0.000 description 2
- 239000003814 drug Substances 0.000 description 2
- 238000002595 magnetic resonance imaging Methods 0.000 description 2
- 208000024891 symptom Diseases 0.000 description 2
- 238000011282 treatment Methods 0.000 description 2
- PGOHTUIFYSHAQG-LJSDBVFPSA-N (2S)-6-amino-2-[[(2S)-5-amino-2-[[(2S)-2-[[(2S)-2-[[(2S)-2-[[(2S)-4-amino-2-[[(2S)-2-[[(2S)-2-[[(2S)-2-[[(2S)-2-[[(2S)-5-amino-2-[[(2S)-5-amino-2-[[(2S)-2-[[(2S)-2-[[(2S)-2-[[(2S,3R)-2-[[(2S)-5-amino-2-[[(2S)-2-[[(2S)-2-[[(2S,3R)-2-[[(2S)-2-[[(2S)-2-[[(2S)-2-[[(2S)-2-[[(2S)-5-amino-2-[[(2S)-1-[(2S,3R)-2-[[(2S)-2-[[(2S)-2-[[(2R)-2-[[(2S)-2-[[(2S)-2-[[2-[[(2S)-2-[[(2S)-2-[[(2S)-2-[[(2S)-1-[(2S)-2-[[(2S)-2-[[(2S)-2-[[(2S)-2-amino-4-methylsulfanylbutanoyl]amino]-3-(1H-indol-3-yl)propanoyl]amino]-5-carbamimidamidopentanoyl]amino]propanoyl]pyrrolidine-2-carbonyl]amino]-3-methylbutanoyl]amino]-4-methylpentanoyl]amino]-4-methylpentanoyl]amino]acetyl]amino]-3-hydroxypropanoyl]amino]-4-methylpentanoyl]amino]-3-sulfanylpropanoyl]amino]-4-methylsulfanylbutanoyl]amino]-5-carbamimidamidopentanoyl]amino]-3-hydroxybutanoyl]pyrrolidine-2-carbonyl]amino]-5-oxopentanoyl]amino]-3-hydroxypropanoyl]amino]-3-hydroxypropanoyl]amino]-3-(1H-imidazol-5-yl)propanoyl]amino]-4-methylpentanoyl]amino]-3-hydroxybutanoyl]amino]-3-(1H-indol-3-yl)propanoyl]amino]-5-carbamimidamidopentanoyl]amino]-5-oxopentanoyl]amino]-3-hydroxybutanoyl]amino]-3-hydroxypropanoyl]amino]-3-carboxypropanoyl]amino]-3-hydroxypropanoyl]amino]-5-oxopentanoyl]amino]-5-oxopentanoyl]amino]-3-phenylpropanoyl]amino]-5-carbamimidamidopentanoyl]amino]-3-methylbutanoyl]amino]-4-methylpentanoyl]amino]-4-oxobutanoyl]amino]-5-carbamimidamidopentanoyl]amino]-3-(1H-indol-3-yl)propanoyl]amino]-4-carboxybutanoyl]amino]-5-oxopentanoyl]amino]hexanoic acid Chemical compound CSCC[C@H](N)C(=O)N[C@@H](Cc1c[nH]c2ccccc12)C(=O)N[C@@H](CCCNC(N)=N)C(=O)N[C@@H](C)C(=O)N1CCC[C@H]1C(=O)N[C@@H](C(C)C)C(=O)N[C@@H](CC(C)C)C(=O)N[C@@H](CC(C)C)C(=O)NCC(=O)N[C@@H](CO)C(=O)N[C@@H](CC(C)C)C(=O)N[C@@H](CS)C(=O)N[C@@H](CCSC)C(=O)N[C@@H](CCCNC(N)=N)C(=O)N[C@@H]([C@@H](C)O)C(=O)N1CCC[C@H]1C(=O)N[C@@H](CCC(N)=O)C(=O)N[C@@H](CO)C(=O)N[C@@H](CO)C(=O)N[C@@H](Cc1cnc[nH]1)C(=O)N[C@@H](CC(C)C)C(=O)N[C@@H]([C@@H](C)O)C(=O)N[C@@H](Cc1c[nH]c2ccccc12)C(=O)N[C@@H](CCCNC(N)=N)C(=O)N[C@@H](CCC(N)=O)C(=O)N[C@@H]([C@@H](C)O)C(=O)N[C@@H](CO)C(=O)N[C@@H](CC(O)=O)C(=O)N[C@@H](CO)C(=O)N[C@@H](CCC(N)=O)C(=O)N[C@@H](CCC(N)=O)C(=O)N[C@@H](Cc1ccccc1)C(=O)N[C@@H](CCCNC(N)=N)C(=O)N[C@@H](C(C)C)C(=O)N[C@@H](CC(C)C)C(=O)N[C@@H](CC(N)=O)C(=O)N[C@@H](CCCNC(N)=N)C(=O)N[C@@H](Cc1c[nH]c2ccccc12)C(=O)N[C@@H](CCC(O)=O)C(=O)N[C@@H](CCC(N)=O)C(=O)N[C@@H](CCCCN)C(O)=O PGOHTUIFYSHAQG-LJSDBVFPSA-N 0.000 description 1
- 208000005156 Dehydration Diseases 0.000 description 1
- 229940121710 HMGCoA reductase inhibitor Drugs 0.000 description 1
- 101000596041 Homo sapiens Plastin-1 Proteins 0.000 description 1
- 206010020772 Hypertension Diseases 0.000 description 1
- 206010028980 Neoplasm Diseases 0.000 description 1
- 102100035181 Plastin-1 Human genes 0.000 description 1
- 108010094028 Prothrombin Proteins 0.000 description 1
- 102100027378 Prothrombin Human genes 0.000 description 1
- 238000001237 Raman spectrum Methods 0.000 description 1
- 208000001647 Renal Insufficiency Diseases 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 108090000190 Thrombin Proteins 0.000 description 1
- 108010000499 Thromboplastin Proteins 0.000 description 1
- 102000002262 Thromboplastin Human genes 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 239000012491 analyte Substances 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 238000009534 blood test Methods 0.000 description 1
- 210000004204 blood vessel Anatomy 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 230000018044 dehydration Effects 0.000 description 1
- 238000006297 dehydration reaction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000002189 fluorescence spectrum Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 238000009532 heart rate measurement Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 201000006370 kidney failure Diseases 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 150000003071 polychlorinated biphenyls Chemical class 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000002600 positron emission tomography Methods 0.000 description 1
- 229940039716 prothrombin Drugs 0.000 description 1
- 230000029058 respiratory gaseous exchange Effects 0.000 description 1
- 210000003296 saliva Anatomy 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000007920 subcutaneous administration Methods 0.000 description 1
- 238000002560 therapeutic procedure Methods 0.000 description 1
- 229960004072 thrombin Drugs 0.000 description 1
- 238000003325 tomography Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000002604 ultrasonography Methods 0.000 description 1
- 210000002700 urine Anatomy 0.000 description 1
- 210000000707 wrist Anatomy 0.000 description 1
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
Definitions
- One or more aspects of embodiments according to the present disclosure relate to health monitoring, and more particularly to a system and method for health state estimation using machine learning.
- Related art methods for monitoring indicators of patient health may have various drawbacks.
- a healthcare provider may receive new information regarding the health of a patient only relatively rarely, and, in some instances, as a result of an adverse change in the health of a patient, resulting in perceptible symptoms. Diagnosing and treating a health condition only after the patient experiences symptoms may result in potentially needless patient suffering, and, in some cases, increased treatment challenges.
- a method including: receiving a first measurement of a subject, the first measurement being a first tissue spectrum of the subject; and generating, using a machine learning inference process based on the first measurement, an estimate of an aspect of the health state of the subject.
- the aspect of the health state of the subject is a concentration of a chemical constituent of a tissue of the subject.
- the chemical constituent is a substance selected from the group consisting of glucose, cortisol, cholesterol, lactate, ethanol, and water.
- the aspect of the health state of the subject is the presence of a medical condition.
- the machine learning inference process is an inference process of a machine learning model selected from the group consisting of neural networks, support vector machines, partial least squares algorithms, decision trees, and combinations thereof.
- the method further includes receiving a second measurement of the subject, wherein the machine learning inference process is further based on the second measurement.
- the second measurement is a second tissue spectrum of the subject.
- the first tissue spectrum and the second tissue spectrum are obtained at different points in time.
- the first tissue spectrum and the second tissue spectrum are obtained at different locations on the body of the subject.
- the second measurement of the subject is a result of a chemical analysis of a sample from the subject.
- the second measurement of the subject is an image of a portion of the subject.
- the method further includes performing a machine learning training process to generate a trained state, wherein the machine learning training process is based on a plurality of measurements of one or more training subjects; and the machine learning inference process is further based on the trained state.
- the machine learning training process is further based on a plurality of target variables, each of the target variables corresponding to a respective measurement of the plurality of measurements.
- the machine learning training process is further based on a measurement of the subject, and a target variable corresponding to the measurement of the subject. [0019] In some embodiments, the machine learning training process comprises clustering.
- the machine learning training process comprises dimensionality reduction.
- a method including: receiving a first measurement of a subject; receiving a second measurement of the subject, obtained after the first measurement; and generating, using a machine learning inference process based on the first measurement and on the second measurement, an estimate of an aspect of the health state of the subject.
- the aspect of the health state of the subject is a concentration of a chemical constituent of a tissue of the subject.
- the chemical constituent is a substance selected from the group consisting of glucose, cortisol, cholesterol, lactate, ethanol, and water.
- the aspect of the health state of the subject is the presence of a medical condition.
- the first measurement is a first tissue spectrum of the subject
- the second measurement is a second tissue spectrum of the subject.
- the method further includes receiving a third measurement of the subject, wherein the machine learning inference process is further based on the third measurement.
- the third measurement is a third tissue spectrum of the subject, and the first tissue spectrum and the third tissue spectrum are obtained at different locations on the body of the subject.
- the first measurement of the subject is a result of a chemical analysis of a first sample from the subject
- the second measurement of the subject is a result of a chemical analysis of a second sample from the subject.
- the first measurement of the subject is a first image of a portion of the subject
- the second measurement of the subject is a second image of a portion of the subject.
- the method further includes performing a machine learning training process to generate a trained state, wherein the machine learning training process is based on a plurality of measurements of one or more training subjects; and the machine learning inference process is further based on the trained state.
- the machine learning training process is further based on a plurality of target variables, each of the target variables corresponding to a respective measurement of the plurality of measurements.
- the machine learning training process comprises clustering.
- the machine learning training process comprises dimensionality reduction.
- the machine learning inference process is an inference process of a machine learning model selected from the group consisting of neural networks, support vector machines, partial least squares algorithms, decision trees, and combinations thereof.
- the machine learning inference process is an inference process of a recurrent neural network.
- a system including: a processing circuit, the processing circuit being configured to: receive a first measurement of a subject, the first measurement being a first tissue spectrum of the subject; and generate, using a machine learning inference process based on the first measurement, an estimate of an aspect of the health state of the subject.
- the aspect of the health state of the subject is a concentration of a chemical constituent of a tissue of the subject.
- the aspect of the health state of the subject is the presence of a medical condition.
- the processing circuit is further configured to receive a second measurement of a subject, the second measurement being a second tissue spectrum of the subject, obtained later than the first measurement; and the generating of the estimate of the aspect of the health state of the subject includes using a machine learning inference process further based on the second measurement.
- the system further includes a portable spectrophotometer, configured to obtain a plurality of tissue spectra including the first tissue spectrum.
- FIG. 1 is a flow chart of a method for training, according to an embodiment of the present disclosure
- FIG. 2 is a flow chart of a method for inference, according to an embodiment of the present disclosure.
- FIG. 3 is a block diagram of a system including a central server and a plurality of sensors, according to an embodiment of the present disclosure.
- a physician may observe a patient to make diagnoses and prescribe treatments. These observations can be costly in terms of time and resources, and they may be invasive, leading to patient discomfort and potentially causing the patient to be reluctant to participate. As a result, such observations may be carried out infrequently, e.g., once per year, especially if the patient has no reason to believe that they are sick. In part because of this paucity of observations, disease states may present themselves once patient illnesses are advanced and more difficult to treat.
- a biometric sensor e.g., a wearable, non-invasive sensor
- a suitable analysis platform e.g., a cloud-based analysis platform
- the analysis platform may be a suitable machine learning model, which may be trained to infer a health outcome (e.g., the presence of a medical condition, such as diabetes) from measurements of a subject (e.g., a patient).
- the training may involve receiving measurements, along with reports of one or more health outcomes, from a plurality of subjects (or “training subjects”), and training the machine learning model to infer, from other measurements, the presence (or absence) of the health outcome.
- the machine learning model may receive measurements of each of a plurality of training subjects who may also report whether or not they have diabetes.
- the measurements may be generated by a wearable spectrophotometer; such measurements may be referred to as “tissue spectra” of the subject or subjects, as discussed in further detail below.
- the tissue spectra may be obtained by reflectance from or transmission though the tissue. Such spectra may be absorption spectra and may be diffuse reflectance absorption spectra. Furthermore, the tissue spectra may be one or more of a variety of spectral types including fluorescence spectra and Raman spectra.
- a measurement from a wearable spectrophotometer may be used as an input observation.
- Training data including a set of such input observations, and a corresponding set of target variables, may be used to train a machine learning model, such as a neural network.
- this training, or “learning” phase may be supervised training in which the training data set 105 includes, for each of a plurality of training subjects (e.g., patients), an input observation, which may be a set of measurements (e.g., one or more tissue spectra) X, and a set of target variables, which may be data Y regarding one or more aspects of a subject’s medical state.
- a “target variable” is one or more numbers, used to train a machine learning model, each representing an aspect of the health state (the aspect being an aspect to be estimated, during inference, by the machine learning model) of the subject.
- An “estimate” is the output of a machine learning model when the machine learning model receives an input observation.
- An estimate may be a single number or a set of numbers.
- an “input observation” is one or more numbers that are used as input to a machine learning model.
- the “health state” of the subject is the entire physical state of the subject’s body.
- the medical state may have various aspects including, for example, the subject’s height and weight, the subject’s blood pressure, temperature, and heart rate, the subject’s blood chemistry (e.g., the concentrations, in the blood, of glucose, cortisol, cholesterol, lactate, or ethanol, or a medication or therapy such as a statin), other characteristics of the blood (e.g., the prothrombin time, activated partial thromboplastin time, or thrombin time), and the presence of any medical conditions (e.g., (i) whether or not the subject has hypertension, cancer, dehydration or diabetes, or is suffering from a heart attack, kidney failure, sepsis or stroke, or (ii) the severity or likelihood of any such condition).
- the subject s height and weight
- the subject’s blood chemistry e.g., the concentrations, in the blood, of glucose, cortisol, cholesterol, lactate, or ethanol, or a medication or therapy such as
- the subject may be a human subject or an animal.
- analogous methods as those disclosed herein for human or animal subjects may be employed to estimate the health state of another living organism, e.g., a plant, or to estimate the state of another object (e.g., the safety and quality of a food product in a factory for making such products).
- the estimate may consist of one or more numbers.
- the estimate may be a single number (e.g., representing the blood glucose level, e.g., in units of mg/dL).
- the presence of these three conditions may be represented as a vector of three numbers, the first of which, for example, may be (i) a number that is zero or one, depending on whether the machine learning model estimates that the subject is suffering from a heart attack, or (ii) a number between zero and one representing the estimated severity of a heart attack (with zero corresponding to the absence of a heart attack), or (iii) a number between zero and one representing the estimated likelihood that the subject is suffering from a heart attack (with, in this case also, zero corresponding to the absence of a heart attack).
- a “medical condition” is an aspect of the medical state, and means a respect in which the medical state of the subject may differ from that of a reference subject, where the reference subject may be, for example, an average or typical subject, or a hypothetical perfectly healthy subject, or the state of the subject at a previous point in time.
- a medical condition may be, but need not be, disadvantageous; for example, an unusually high aerobic capacity is a medical condition as that term is used herein.
- the training may be conducted, using a suitable cost function, at 110 (in FIG. 1 ), to cause the machine learning model to associate certain combinations of measurement values (which may be considered to be, or arranged as, a measurement vector or array) with certain aspects of the subject’s medical state.
- the training data set may include, for each of the training subjects, a tissue spectrum as the input observation X, and a blood glucose level (separately measured, e.g., by chemical analysis of a blood sample) as the aspect, of the medical states of the training subjects, that may form the set of target variables Y.
- the cost function may be a function of the difference between (i) an estimate, of the blood glucose level, produced by the machine learning model and (ii) the corresponding target variable, e.g., the separately measured blood glucose level.
- the machine learning model may be capable of estimating the aspects of the medical state of the subject with which it was trained (e.g., the blood glucose level) from the measurement of a subject (e.g., the tissue spectrum of the subject).
- the machine learning model may be capable of estimating aspects of the medical state of a subject at a later time, in other words, of predicting a future medical state.
- the training may result in the adjustment of parameters in the machine learning model so as to reduce or minimize the cost function.
- “minimizing” the cost function means adjusting the parameters (e.g., iteratively) until a termination condition is satisfied.
- the termination condition may be satisfied, e.g., if the number of iterations that have been performed has reached a threshold, or if the improvement (reduction) in the cost function, per iteration, has fallen below a threshold.
- the machine learning model is a neural network
- the training may result in the adjustment of weights in the neural network).
- the set of values that these parameters take after training may be referred to as the “trained state” of the machine learning model.
- the parameter adjustment process can be carried out using stochastic gradient descent, in the case of a neural network model.
- the trained state of the machine learning model may be deployed to one or more inference servers, if these inference servers are separate from the central server (in some embodiments the central server 305 is the inference server or one of the inference servers). Equivalently, a plurality of inference servers, connected to each other or each connected to a central server may be considered to be a single distributed server.
- the trained machine learning model may, at 205, receive an input observation (e.g., sensor measurements X), and generate, at 210, an estimate, e.g., estimates Y’ of aspects of the subject’s medical state, from the input observation.
- the inference server may estimate the subject’s blood glucose level from a tissue spectrum of the subject. A portion of the estimate may then, at 215, be sent to the subject, or to the subject’s health care providers.
- training may continue after the inference phase has begun, e.g., the parameters of the machine learning model may periodically or continuously be adjusted, to minimize the cost function, as new training data become available (e.g., from the training subjects, or from the subject (as discussed in further detail below)).
- the parameters of the machine learning model may be, e.g., global parameters (based on “global” training data from all training subjects), or they may be subject-specific parameters (or some parameters may be global, and some may be subject-specific).
- the parameters may be stored in the inference server, or in a device (e.g., in a sensor, or in an interface device such as a mobile device with a display that is configured to communicate with a sensor and with the inference server) associated with a sensor or with a subject.
- subject-specific training data may be obtained. For example, if a subject obtains independent confirmation (or refutation) of an estimate (e.g., the presence of diabetes) generated by the machine learning model, the subject’s input observation (e.g., the subject’s tissue spectrum) and the subject’s target variable (e.g., the independently determined presence or absence of diabetes) may be used as subjectspecific training data.
- the subject-specific data may be (i) added to the (global) training data set, or (ii) used to perform subject-specific training.
- Subject-specific training may involve, e.g., using a set of parameter values obtained by training the machine learning model with the global training data set as a set of initial values, and then adjusting some or all of the parameters (with, in some embodiments, some parameters being frozen) to minimize the cost function for (i) the global training data set, augmented with the subject-specific training data, or (ii) the subject-specific training data only.
- the measurement or measurements upon which the estimate or estimate generated by the machine learning model are based may be high-dimensional data which may be, or include, measurements made by a wide range of methods for obtaining information about the subject, including, for example (as mentioned above), one or more tissue spectra (e.g., obtained at one or more locations on the subject’s body or obtained at one or more points in time), chemical analyses of samples (e.g., blood, saliva, urine or breath samples), physical measurements (e.g., height, weight, age, skin tone, temperature, strength, heart rate, blood pressure, or breathing rate), text (e.g.
- tissue spectra e.g., obtained at one or more locations on the subject’s body or obtained at one or more points in time
- chemical analyses of samples e.g., blood, saliva, urine or breath samples
- physical measurements e.g., height, weight, age, skin tone, temperature, strength, heart rate, blood pressure, or breathing rate
- text e.g.
- a measurement of a subject may be a measurement of any aspect (or any set of aspects) of the medical state of the subject, and any such measurements may be part of the input observation.
- a tissue spectrum may be a portion of the input observation in a machine learning model used to estimate a blood glucose level from tissue spectra, as explained above.
- a tissue spectrum may be used to infer an aspect of the subject’s medical state, and these derived data may be part of the target variable used for training (e.g., a blood alcohol level inferred from a tissue spectrum may be used to train a machine learning model to estimate blood alcohol level from other measurements of the subject).
- the target variables used for training may include opinions of human experts (e.g., opinions regarding, the presence of a certain medical condition in a subject).
- a sensor may be a wearable spectrophotometer which (i) illuminates the skin of a subject with a probe beam including light of different wavelengths and (ii) measures light returning to the spectrophotometer, e.g., after scattering from subcutaneous tissue of the subject.
- the spectrophotometer may produce a probe beam with light at, e.g., each of 100 different wavelengths in turn, and, for each wavelength, measure the intensity of the returning light.
- the wavelengths may be near infrared (NIR) wavelengths, which may be in the range from 780 nm to 2500 nm, or, e.g., from 1200 nm to 1800 nm.
- NIR near infrared
- the ratio, as a function of wavelength, of (i) the intensity of the light returning to the spectrophotometer to (ii) the power of the probe beam, may depend on, and therefore provide information about, the subcutaneous tissue.
- a ratio, as a function of wavelength may (as mentioned above) be referred to as a “spectrum” or “tissue spectrum” of the subject.
- Such information may include, for example, information about the chemical composition of the subcutaneous tissue.
- tissue refers to any biological component of the subject, including, for example, the skin of the subject, and blood in subcutaneous blood vessels of the subject.
- a subject may be fitted with more than one such sensor; e.g., the subject may wear one sensor on the wrist and another on the chest.
- the tissue spectrum may be represented in any of a number of ways.
- a tissue spectrum may be represented as a vector, or array, of numbers, one for each wavelength of operation of the spectrophotometer, each number being the ratio at the respective wavelength, of (i) the intensity of the light returning to the spectrophotometer to (ii) the power of the probe beam.
- the tissue spectrum may be represented otherwise, e.g., using a representation that is independent of the wavelengths (and of the number or wavelengths) of operation of the spectrophotometer.
- the tissue spectrum may be approximated by a continuous function, and converted to a spectrophotometer-independent representation by resampling the continuous function at another set of wavelengths.
- the continuous function may be defined by a set of parameters (e.g., polynomial coefficients, if the continuous function is, e.g., a polynomial or a cubic spline, or a set of peak magnitudes and peak widths if the continuous function is or includes a superposition of one or more peaks, each peak being a Lambertian function), and the set of parameters may be employed as a spectrophotometer-independent representation of the tissue spectrum.
- the use of a spectrophotometer-independent representation may be of use, for example, if different spectrophotometers are used by different subjects or at different times.
- the input observation may include a time series of similarly- obtained measurements, e.g., a series of tissue spectra.
- a time series of other measurements may be, or be part of, the input observation used for performing an inference process.
- the input observation may include a time series of heart rate measurements, a time series of blood test results, or a series of MRI images. If a time series of tissue spectra is used, the spectra may be obtained at a rate, for example, of 10 or 20 Hz, so that variations with the cardiac cycle may be resolvable in the data.
- Such a time series of tissue spectra may form a two-dimensional array, with one dimension corresponding to time and the other dimension corresponding to wavelength.
- Various types of machine learning models may be used, including neural networks (e.g., convolutional neural networks), support vector machines, partial least squares algorithms, and decision trees. If the input observation is or includes a time series of measurements, a transformer or a recurrent neural network may be used as the machine learning model. For example, in the case of a spectrophotometer sensor with 194 wavelengths, a 5-layered 1 -dimensional convolutional neural network may be used to estimate the level of a particular blood analyte (e.g., to estimate the level of a chemical component of the blood).
- the input layer may have 2 feature maps that are 9x1 , then 2 feature maps that are 7x1 , then 4 feature maps that are 7x1 , then 8 feature maps that are then 5x1 , then 12 feature maps that are 3x1 , followed by a fully-connected layer with a single output, where each convolution operation is followed by a rectified linear operation, then a batch normalization step, then an average pooling layer.
- a partial least squares (PLS) model may be used as the machine learning model.
- the matrix X can be constructed by arranging n rows of 194-dimensional observations into a n by 194 matrix of predictors
- the matrix Y can be constructed by arranging n rows of p dimensional target variables, then applying the PLS1 algorithm to estimate factor and loading matrices and error matrices given X and Y.
- a gradient-boosted decision tree model may be used as the machine learning model.
- X and Y matrices may be built as in the PLS case, and an ensemble of 10 decision trees with a maximum depth of 5 may be built using the XGBoost algorithm with either a linear or logistic objective function for regression or classification problems, respectively.
- a regularized linear regression model such as LASSO or ridge regression, may be used as a machine learning model.
- LASSO LASSO or ridge regression
- a machine learning model may be used that carries out unsupervised learning.
- unsupervised learning the machine learning algorithm uses only the X matrix as input.
- clustering where the algorithm uses only the X matrix and outputs an estimated “class” for each observation (in this example, each row) of the X matrix.
- Another type of unsupervised learning is dimensionality reduction, which takes the observation matrix X and solves for a transformation T that maps each observation into a lower dimensional space while preserving specified properties of the data.
- PCA principal components analysis
- PCA principal components analysis
- N may be 10
- one may project the data to an N-dimensional subspace that can be used as input to a downstream machine learning model.
- a central server 310 is connected to a plurality of sensors 310 (e.g., portable or wearable spectrophotometers, as discussed above), each of which takes, from time to time, or substantially continuously (e.g., with a sample rate of 10 Hz or 20 Hz), measurements of a respective subject (or, in some cases, several of which may be taking measurements of the same subject).
- the sensors 310 relay the measurements to the central server 305.
- target variable data are also sent to the central server 305.
- the target variable data may be sent separately, or over shared data paths.
- each sensor may be a wearable spectrophotometer worn by a respective subject.
- Each subject may also carry a portable device such as a mobile phone, which may (i) receive the measurements via a local (e.g., WiFi or Bluetooth) data connection and relay the measurements to the central server 305 through the Internet (e.g., via a mobile telephone network).
- the subject may also obtain target variable data (e.g., using an additional sensor) and transmit the data to the central server 305 through the same connection between the mobile device and the central server 305.
- target variable data e.g., using an additional sensor
- the subject may have a blood glucose measuring device that displays blood glucose level measurements to the subject, who then keys these measurements into the mobile device (or the blood glucose measuring device may communicate these measurements directly with the mobile device through a local connection) for transmission to the central server 305.
- the inference process may also employ the configuration of FIG. 3.
- an interference server (which may be the same central server 305, or a separate server) may similarly receive sensor data from a plurality of sensors.
- the inference server may generate estimates of aspects of the subject’s medical state. These estimates may be, for example, (i) sent back to the subject, (ii) sent to the subject’s healthcare provider, or (iii) compared to prescribed ranges, and, if they fall within such ranges, used to trigger additional steps, such as alerts (e.g., a warning to the subject if an estimated blood glucose level is dangerously high), medication, or other interventions.
- a portion of the data generated by the sensor may be replaced with a lower volume of processed data. For example, outliers, or data that are of relatively low value for inference (and for training) may be dropped, or averaged with other data. For example, if blood glucose level is to be estimated, wavelengths at which the spectrum varies relatively little as a function of blood glucose level may be omitted from the input observation. As another example, if the raw data obtained by the sensor vary with the cardiac cycle, then the raw data may be fit to the cardiac cycle, and lower-rate average data may be transmitted to the central server 305.
- the “presence of” a substance or a medical condition means whether or not the substance or medical condition is present, or the extent to which it is present (e.g., the severity of a medical condition or the concentration of a substance in the blood), or the likelihood of the substance or medical condition being present.
- “estimating the presence of” something means estimating whether or not the thing is present, or estimating the extent to which the thing is present, or estimating the likelihood of the thing being present.
- a portion of something means “at least some of’ the thing, and as such may mean less than all of, or all of, the thing.
- “a portion of” a thing includes the entire thing as a special case, i.e. , the entire thing is an example of a portion of the thing.
- the word “or” is inclusive, so that, for example, “A or B” means any one of (i) A, (ii) B, and (iii) A and B.
- array refers to an ordered set of numbers regardless of how stored (e.g., whether stored in consecutive memory locations, in a linked list, or in another manner).
- the central server and each of the one or more inference servers may be, or include a processing circuit configured to perform portions of the methods disclosed herein.
- processing circuit is used herein to mean any combination of hardware, firmware, and software, employed to process data or digital signals.
- Processing circuit hardware may include, for example, application specific integrated circuits (ASICs), general purpose or special purpose central processing units (CPUs), digital signal processors (DSPs), graphics processing units (GPUs), and programmable logic devices such as field programmable gate arrays (FPGAs).
- ASICs application specific integrated circuits
- CPUs general purpose or special purpose central processing units
- DSPs digital signal processors
- GPUs graphics processing units
- FPGAs programmable logic devices
- each function is performed either by hardware configured, i.e., hard-wired, to perform that function, or by more general- purpose hardware, such as a CPU, configured to execute instructions stored in a non- transitory storage medium.
- a processing circuit may be fabricated on a single printed circuit board (PCB) or distributed over several interconnected PCBs.
- a processing circuit may contain other processing circuits; for example, a processing circuit may include two processing circuits, an FPGA and a CPU, interconnected on a PCB.
- a method e.g., an adjustment
- a first quantity e.g., a first variable
- a second quantity e.g., a second variable
- the second quantity is an input to the method or influences the first quantity
- the second quantity may be an input (e.g., the only input, or one of several inputs) to a function that calculates the first quantity, or the first quantity may be equal to the second quantity, or the first quantity may be the same as (e.g., stored at the same location or locations in memory as) the second quantity.
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
Description
Claims
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202280059645.5A CN117897775A (en) | 2021-09-01 | 2022-09-01 | Health estimation using machine learning |
EP22773158.5A EP4396840A1 (en) | 2021-09-01 | 2022-09-01 | Health state estimation using machine learning |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US202163239857P | 2021-09-01 | 2021-09-01 | |
US63/239,857 | 2021-09-01 | ||
US17/823,505 | 2022-08-30 | ||
US17/823,505 US20230215565A1 (en) | 2021-09-01 | 2022-08-30 | Health state estimation using machine learning |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023031355A1 true WO2023031355A1 (en) | 2023-03-09 |
Family
ID=83362624
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/EP2022/074380 WO2023031355A1 (en) | 2021-09-01 | 2022-09-01 | Health state estimation using machine learning |
Country Status (2)
Country | Link |
---|---|
EP (1) | EP4396840A1 (en) |
WO (1) | WO2023031355A1 (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019161411A1 (en) * | 2018-02-17 | 2019-08-22 | Sanmina Corporation | System and method for obtaining health data using a neural network |
WO2020230123A1 (en) * | 2019-05-12 | 2020-11-19 | Makesense Digital Health Technologies Ltd. | A system and a method for health and diet management and nutritional monitoring |
US20200375549A1 (en) * | 2019-05-31 | 2020-12-03 | Informed Data Systems Inc. D/B/A One Drop | Systems for biomonitoring and blood glucose forecasting, and associated methods |
US20210212606A1 (en) * | 2020-01-15 | 2021-07-15 | Bao Tran | Glucose management |
-
2022
- 2022-09-01 WO PCT/EP2022/074380 patent/WO2023031355A1/en active Application Filing
- 2022-09-01 EP EP22773158.5A patent/EP4396840A1/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019161411A1 (en) * | 2018-02-17 | 2019-08-22 | Sanmina Corporation | System and method for obtaining health data using a neural network |
WO2020230123A1 (en) * | 2019-05-12 | 2020-11-19 | Makesense Digital Health Technologies Ltd. | A system and a method for health and diet management and nutritional monitoring |
US20200375549A1 (en) * | 2019-05-31 | 2020-12-03 | Informed Data Systems Inc. D/B/A One Drop | Systems for biomonitoring and blood glucose forecasting, and associated methods |
US20210212606A1 (en) * | 2020-01-15 | 2021-07-15 | Bao Tran | Glucose management |
Also Published As
Publication number | Publication date |
---|---|
EP4396840A1 (en) | 2024-07-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10952682B2 (en) | System and method of a biosensor for detection of health parameters | |
Jalali et al. | Prediction of periventricular leukomalacia in neonates after cardiac surgery using machine learning algorithms | |
CN111508607B (en) | Obesity prediction system based on BP neural network | |
WO2019196076A1 (en) | Systems and methods for determining blood pressure of subject | |
Ahmed et al. | Performance of artificial intelligence models in estimating blood glucose level among diabetic patients using non-invasive wearable device data | |
El-Dahshan et al. | ExHyptNet: An explainable diagnosis of hypertension using EfficientNet with PPG signals | |
CN115330670A (en) | Method and system for simultaneous classification and regression of clinical data | |
Bodenes et al. | Early heart rate variability evaluation enables to predict ICU patients’ outcome | |
Man et al. | Improving non-invasive hemoglobin measurement accuracy using nonparametric models | |
Padmavilochanan et al. | Personalized diabetes monitoring platform leveraging IoMT and AI for non-invasive estimation | |
WO2022216220A1 (en) | Method and system for personalized prediction of infection and sepsis | |
Liu et al. | PbFG: Physique-based fuzzy granular modeling for non-invasive blood glucose monitoring | |
US20230215565A1 (en) | Health state estimation using machine learning | |
EP4396840A1 (en) | Health state estimation using machine learning | |
CN117897775A (en) | Health estimation using machine learning | |
CN115458163A (en) | Characteristic construction method, system, device and medium based on metabolic synthesis index | |
US20220400960A1 (en) | Autoregulation monitoring using deep learning | |
Valero et al. | Pervasive glucose monitoring: A non-invasive approach based on near-infrared spectroscopy | |
Duc et al. | Non-invasive in vivo type 2 diabetes Mellitus diagnosis using Raman Spectroscopy in Combination with Machine Learning | |
Raju et al. | DNN-BP: a novel framework for cuffless blood pressure measurement from optimal PPG features using deep learning model | |
Byfield et al. | Equine Life Stage Classification from Photoplethysmography Data by an Explainable Echo State Network | |
US20240321447A1 (en) | Method and System for Personalized Prediction of Infection and Sepsis | |
US20240108316A1 (en) | Processing an Arterial Doppler Ultrasound Waveform | |
US11839471B2 (en) | Autoregulation monitoring using deep learning | |
Chen | NON-INVASIVE BLOOD GLUCOSE MEASUREMENT WITH MID-INFRARED SIGNAL BY LEARNING SCHEMES |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22773158 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202280059645.5 Country of ref document: CN |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2022773158 Country of ref document: EP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2022773158 Country of ref document: EP Effective date: 20240402 |