Predicting body mass index and isometric leg strength using soft tissue distributions from computed tomography scans

This paper describes the interconnections and predictive value between Body Mass Index (BMI), Isometric Leg Strength (ISO) and soft tissue distribution from mid-thigh Computed Tomography (CT) scans using Machine Learning (ML) regression and classification algorithms. A novel methodology for soft tissue patient specific CT profile called Nonlinear Trimodal Regression Analysis (NTRA) was developed using radiodensitomentric distribution from a CT scan. This method defines 11 parameters used as input features for Tree-Based ML algorithms in order to apply regression and classification on BMI and ISO. K_fold Cross-Validation with k = 10 is applied to obtain several models to choose the best one using the higher coefficient of determination (R2) as an evaluator of the quality of regression prediction. Following this, BMI and ISO are divided into 3 and 5 classes and the same methodology is used to classify them. For this analysis, an accuracy parameter is calculated to evaluate the quality of the results. The max R2 is 88.9 for the BMI and it is obtained using the Gradient-Boosting Algorithm. The best accuracy was 76.1 for 3 classes and 73.1 for 5 classes. The best results obtained for ISO are R2 = 66.5 and an accuracy of 65.5 for the 3 classes classification. Furthermore, the connective tissue assumes high importance in the prediction process. In this methodological study the feasibility of a ML approach was tested with good results, in order to show a novel approach to study the correlation between physiology parameters and imaging.