Hybrid ML Biophysical Model for Hidden Metric Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Biophysical models used to simulate biological systems are limited by the large number of input parameters that are not easily measurable, making it difficult to accurately predict patient health states and disease progression due to the lack of measurable hidden biological metrics.
Innovation Solution
A hybrid machine learning approach is employed, combining mechanistic models with machine learning models to infer and predict hidden biophysical metrics from clinical data, using a generative model to transform random variables into mechanistic model parameters and a surrogate model to replicate the logic of the mechanistic model, enabling the prediction of patient classifications based on non-invasively measurable data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If biophysical models use a large number of input parameters to accurately simulate biological systems, then prediction accuracy improves, but measurement difficulty increases because many parameters are not easily measurable
Solution Approach 1:
The patent introduces machine learning models as intermediary components that bridge the gap between easily measurable clinical data and the hidden biophysical parameters needed for accurate biophysical modeling. The ML models infer unmeasurable parameters from observable clinical measurements, enabling the biophysical model to function with complete input data without requiring direct measurement of all parameters
Solution Approach 2:
The patent replaces the direct measurement approach (mechanical/physical measurement systems) with a computational inference approach using machine learning algorithms. Instead of physically measuring hard-to-obtain biophysical parameters, the system uses ML models to compute these parameters from easily measurable clinical data, substituting physical measurement with computational prediction
2Measurement precision
If biophysical models include many input parameters to capture complex biological systems, then model accuracy improves, but model complexity increases making the models harder to operate and implement
Solution Approach 1:
The patent segments the overall modeling task into distinct components: a machine learning model for inferring hidden parameters and a biophysical model for simulating biological systems. This segmentation allows each component to specialize in specific functions, reducing the operational complexity of the overall system while maintaining comprehensive modeling capability
Solution Approach 2:
The machine learning model serves as an intermediary layer that preprocesses clinical data and generates inferred biophysical parameters before they enter the main biophysical simulation model. This intermediary processing simplifies the input requirements of the biophysical model and makes the overall system more manageable despite the large number of parameters involved
Data Source
AI summary
Mechanisms are provided for training a hybrid machine learning (ML) computer model to simulate a biophysical system of a patient and predict patient classifications based on results of simulating the biophysical system. A mechanistic model is executed to generate a training dataset. A surrogate ML model is trained to replicate logic of the mechanistic computer model and generate patient feature outputs based on surrogate ML model input parameters. A transformation ML model is trained to transform patient feature outputs of the surrogate ML model into a distribution of patient features. A generative ML model is trained to encode samples from a uniform distribution of input patient data into mechanistic model parameter inputs that are coherent to the target distribution of patient features and are input to the surrogate ML model. Input patient data for a patient is processed through the ML models to predict a patient classification for the patient.