Machine Learning Methods for Predicting Tumor Volume in Rats after Termination of Complex Treatment with a Varying Dose of Cyclophosphamide

Predictive determination of tumor volume is a significant task in cancer treatment. The importance of predictive definition is due to its relevance in research and clinical practice. The use of predictive models makes it possible to evaluate the effectiveness of the treatment used in the early stages and decide on its correction in case of unsatisfactory prognosis. The paper investigates models for predictive determination of tumor volume after applying a complex treatment that combines chemotherapy and hyperthermia. Considered machine learning models are ensembles of decision trees, regression based on Gaussian processes, ridge regression, and Bayesian regression. The raw data for machine learning comes from an experiment on laboratory rats. As part of the experiment, all animals received an injection of the fast-growing Walker-256 carcinoma into the left thigh. After the injection, the animals were subjected to complex treatment with cyclophosphamide and hyperthermia. During the study, the rats were divided into three groups. The division into groups depended on the dose of cyclophosphamide received. The determination of informative features is based on the application of greedy and genetic algorithms. The analysis of the selected features with their use in various types of models. The projections of remission and the depth of remission were calculated based on the tumor volume\u0027s predicted values. The applicability of several investigated models for predictive assessment of the therapeutic effect of combined treatment has been shown using Student’s and Friedman\u0027s statistical tests.