Assessing the effect of an antitumor drug from preclinical models through advanced data analysis

Abstract

Glioblastoma multiforme (GBM) is a highly aggressive brain tumor that requires early and accurate diagnosis to improve treatment options and patient prognosis. In this work, an automated classification system of treated/responding and untreated GBM preclinical subjects is proposed by using magnetic resonance imaging (MRI) of tumor-bearing mice. By refining this classification, our aim was to improve the precision of treatment follow-up, which still is a current challenge in clinical pipeline for this disease. For this, advanced image processing and deep learning techniques were used. Specifically, various CNN architectures and transfer learning techniques have been used in order to take advantage of the prior knowledge of pre-trained models. To train and validate the proposed models, a data set of mice with treated and control (untreated) tumors has been used. The results show that the system classifies these cases with great precision. The best performance metrics obtained include accuracy greater than 97\% and area under the ROC curve (AUC) greater than 0.99 on test data sets. This work aims to contribute to the advancement of scientific knowledge in the field of medical imaging and machine learning applied to health. Furthermore, it provides an automated solution with translational potential that could assist medical professionals in the future. The use of preclinical models allow longitudinal and extensive validation, establishing a bridge for potential future application in human patients.