Artificial intelligence-based estimation of body muscle percentage with biomedical signals

Measuring and monitoring the body muscle mass required to maintain a healthy life is essential. Especially, since muscle mass affects the life quality of elderly persons directly, monitoring the muscle mass in these persons has more importance. Since the current methods utilised to measure muscle mass have some disadvantages, it is required practical, reliable and high-tech devices can be used to measure muscle mass. This study aims to develop a low-cost and reliable BMP calculation model using artificial intelligence algorithms and biomedical signals. In the study, 300 Electrocardiography (ECG) signals belonging to the individual were used. Firstly, the ECG signals were filtered, and sub-frequency bands were obtained. From the ECG signal of each individual, the QRS components of this signal at 7 different frequencies were obtained. Thus, 8 signals were acquired for each individual. A total of 200-time domain features were extracted, 25 of which were obtained from each signal. In addition, five demographic features (age, weight, height, BMI, muscle) were added to the model, and 205 features were reached. To enhance the performance, the spearman feature selection algorithm was used. As machine learning algorithms; Decision Trees, Support Vector Regression, Ensemble Decision Trees have been used. The recommended BMP estimation model has the performance values for all individuals MAPE=4,18 (Ensemble Decision Tree), for males MAPE=3,91 (Support Vector Regression) and for females MAPE=4,54 (Support Vector Regression) in this study. Regarding the results of the study, It is thought that ECG-based BMP prediction models can be used. Keywords: Electrocardiography Signal, Machine Learning, Artificial Intelligence, Body Composition, Body Muscle Percentage, Gender-Based Body Muscle Percentage