Predictor of neoplasms and body composition with machine learning models

Abstract

AbstractBackgroundThe tissue microenvironment of neoplastic diseases differs from that of normal cells. Their extracellular matrix, innervation, metabolism, as well as interstitial fluid and intercellular interconnections mark clear physical differences between normal and cancerous cellular ecosystems. Detecting these physical changes early without using diagnostic methods that are harmful and uncomfortable for the patient is a major challenge for the medical-scientific community. Electrical bioimpedance supported by machine learning techniques can provide clues to incipient preneoplastic tissue changes.MethodsIn this study, 7 predictive machine learning models were developed using a database with bioimpedanciometric data from a group of healthy individuals and another group of patients who had or were suffering from cancer at the time of measurement.ResultsTheRandom Forestwas the model that reported the best Accuracy, reaching over 90% of hits.ConclusionsThese results open the door to future research linking changes in body composition and pretumoral tissue environments using machine learning tools.