A novel dual-channel brain tumor detection system for MR images using dynamic and static features with conventional machine learning techniques

Brain tumor detection, at the early stages of its development for the timely cure of the patient, is a challenging task due to its complex nonlinear nature. We propose a dual-channel brain tumor detection (DC-BTD) framework for magnetic resonance imaging scans with optimum false negatives, based on the idea of using D-channel for extremely discriminant dynamic features and S-channel for static features using data normalization, augmentation, and different machine learning (ML) classifiers, namely, support vector machine, k-nearest neighbor, naive Bayes, and XGBoost. The D-channel features are extracted using a proposed Fine-Tuned Convolutional Neural Network (FT-CNN), while the S-channel features are extracted using the histogram of oriented gradients (HOG)- and local binary patterns (LBP)-based operators. The dual-channel data form the hybrid feature space (HFS) gives improved performance using two types of features. Computer experiments have been conducted on publically available brain tumors dataset obtained from Nanfang Hospital, Guangzhou, and Tianjin Medical University General Hospital, China. The finding of the current study shows that the proposed framework for brain-tumor prediction outperforms other contemporary existing methods with the highest generalization performance as 98.70% (accuracy) and 98.56% (F-score) on the same dataset that can lead to better brain tumor detection at an early stage.

Automated summary

In this study, a dual-channel brain tumor detection framework is proposed to improve the detection performance by using dynamic and static features. Computer experiments on a public brain tumor dataset show that the proposed framework outperforms other existing methods with high accuracy and F-score.