An effective flowchart for multimodal brain tumor binary classification with ranked 3D texture features.

Brain tumors have complex structures, and their shape, density, and size can vary widely. Consequently, their accurate classification, which involves identifying features that best describe the tumor data, is challenging. Using classical 2D texture features can yield only limited accuracy. Here, we show that this limitation can be overcome by using 3D feature extraction and ranking methods. Brain tumor images obtained through 3D magnetic resonance imaging were used to classify high-grade and low-grade glioma in the BraTS 2017 dataset. From the dataset, texture properties for each of the four phases (i.e., FLAIR, T1, T1c, and T2) were extracted using a 3D gray level co-occurrence matrix. Various combinations of brain tumor feature sets were created, and feature ranking methods-Bhattacharyya, entropy, receiver operating characteristic, the t-test, and the Wilcoxon test-were applied to them. Features were classified using gradient boosting, support vector machines (SVMs), and random forest methods. The performance of all combinations was evaluated from the sensitivity, specificity, accuracy, precision, and F-score obtained from twofold, fivefold, and tenfold cross-validation tests. In all experiments, the most effective scheme was that involving the quadruple combination (FLAIR + T1 + T1c + T2) and the entropy feature-ranking method with twofold cross-validation. Notably, the proposed machine-learning-based framework showed remarkable scores of 100% (sensitivity), 97.29% (specificity), 99.30% (accuracy), 99.07% (precision), and 99.53% (F-score) for glioma classification with an SVM. The proposed flowchart reflects a novel brain tumor classification system that competes with the novel methods.