Machine Learning Algorithms Predict Body Mass Index Using Nonlinear Trimodal Regression Analysis from Computed Tomography Scans

In this study Machine Learning supervised regression and classification algorithms are used to predict Body Mass Index (BMI), starting from Computed Tomography scans (CT). From each patient CTs, 11 parameters describing muscle, connective tissue and fat, are extracted creating a patient specific soft tissue profile called Nonlinear Trimodal Regression Analysis (NTRA). Regression and classification are applied in order to predict and classify BMI using Tree-Based algorithms. A proper Train-Test division of the dataset is applied using k_fold Cross-Validation. Various combinations of features are employed with k_fold division in order to obtain the best coefficient of determination (R2) as evaluator of the quality of regression’s prediction. Afterward, BMI is divided into 3 and 5 classes and the same methodology is used to classify it. For this analysis, the accuracy parameter is calculated to evaluate the quality of the results. The max R2 is 0,83 and it is obtained using the 11 NTRA parameters as regressors, k_fold = 16, and the Gradient-Boosting Algorithm. The amplitude of the connective and fat tissue always covers more than 50% of all the feature importance. The best accuracy was 0,80 for 3 classes and 0,74 for 5 classes. The results prove that the 11 NTRA parameters can have a very significant predictive value and the same methodology can be applied in future works to predict other physiological parameters and comorbidities.