Mr Image Processing to Predict Complete Responders by Evaluating the Tumor Regression Grade: A Sensitivity Analysis

Abstract

This work aims to realize a computer-aided method in order to correctly predict and classify complete responders (CRs) patients, with rectal cancer diseases diagnosed and treated with neoadjuvant radiochemotherapy (RCT), employing the tumor regression grade (MR-TRG) estimated by magnetic resonance imaging. The study involved a total of 65 patients and a 3.0 Tesla scanner was employed to perform the magnetic resonance (MR) examinations in order to calculate TRG. The automatic method, by processing and testing patients data, allows to determine the optimum threshold dividing CRs patients from patients that are considered non responders. To automatically determine the best testing rule, a cross-validation statistical analysis was carried out to evaluate the prediction accuracy of the classifier. The algorithm collected the outcomes of the performed cross-validation analysis and the obtained results show the percentages of correct instances and misclassified patients. A sensitivity analysis has also been carried out to study the effect of non-optimum thresholds in the classification procedure. The computer-aided classification of CRs appears to be feasible and it may represent a helpful method to recognize CRs patients, supporting clinicians performing disease prognoses and patient survival expectations in order to provide treatments customization.