ML Model for REE Estimation from CT Scan Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating Resting Energy Expenditure (REE) using imaging techniques, such as MRI and CT scans, suffer from low accuracy and are affected by factors related to specific subjects, particularly in cases of heterogeneous tissue composition due to diseases like muscular dystrophy or hepatic steatosis.
Innovation Solution
A computer-implemented machine learning model is trained using CT scan images to accurately estimate REE parameters. This involves obtaining ground-truth REE parameters through indirect calorimetry or bioimpedanciometry, identifying regions of interest (ROIs) in CT scans, calculating relevant parameters such as radiodensity and volume, and performing supervised training of the model using these parameters and associated ground-truth REE values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If imaging techniques (MRI, CT scans) are used to estimate REE, then quantitative information about tissue volume can be obtained, but measurement precision is low due to manual evaluation and inter-observer dependencies
Solution Approach 1:
The patent replaces manual mechanical evaluation of imaging data with an automated machine learning system. The ML model processes CT scan images to automatically estimate REE, eliminating the need for manual visual inspection and reducing inter-observer variability while maintaining measurement precision.
Solution Approach 2:
The patent creates a computational model that learns from training data to replicate the relationship between tissue composition and REE. This digital copy of the physiological relationship allows automated prediction without requiring repeated manual assessments.
2Productivity
If manual selection of tissues is performed for quantitative analysis, then specific tissue volumes can be measured, but productivity is reduced due to repetitive and time-consuming tasks
Solution Approach 1:
The patent enables the system to automatically identify and segment relevant tissues without human intervention. The machine learning model performs self-service by autonomously processing CT images, extracting features, and generating REE estimates, thereby eliminating repetitive manual tasks and significantly improving productivity.
3Reliability
If whole-body information is used for REE estimation, then comprehensive body composition data is obtained, but harmful factors increase due to high quantity of radiation from CT scans
Solution Approach 1:
The patent extracts only the necessary information for REE estimation from CT scans by focusing on specific regions of interest and key tissue types. This selective approach reduces the need for comprehensive whole-body scanning, thereby minimizing radiation exposure while maintaining sufficient reliability for energy expenditure prediction.
4Measurement precision
If homogeneous tissue composition is assumed for REE calculation, then simplified calculations can be performed, but measurement precision decreases for subjects with heterogeneous tissue composition due to diseases
Solution Approach 1:
The patent applies local quality by analyzing different tissue types separately with their own characteristic properties. The machine learning model identifies and processes various tissues (muscle, fat, organs) with their specific radiodensity and metabolic rates, allowing accurate REE estimation for subjects with heterogeneous tissue composition without requiring complex manual tissue characterization.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A training method of a computer-implemented machine learning model for obtaining a Resting Energy Expenditure (REE) parameter of a subject, and a system to perform such method, are presented, wherein an REE parameter from an experimental test, and information from a CT scan image of the subject are used to train the computer-implemented machine learning model. A method for estimating a Resting Energy Expenditure (REE) parameter of a subject is also presented, using a computer-implemented machine learning model previously trained by the training method described herein.