Towards an Explainable Approach: Hybrid Residual Network and SVM for Automated Brain Tumor Detection and Classification

Abstract

The abnormal growth of brain cells leads to tumor formation, which can be fatal if not detected and treated promptly. Given the complexity of brain tumors, early detection is critical in healthcare. Traditional radiology-based tumor detection is prone to human error and delays. Hence, a computer- assisted method is needed for accurate and efficient diagnosis. With the rapid advancements in the medical science; integrating machine learning, deep learning, artificial intelligence demonstrated great potential in diagnosing diseases and overcome the existing drawbacks while focusing on appropriate treatment plans and improved patient outcomes. A pre-trained model namely Residual Network i.e., ResNet101V2 has been leveraged in the proposed model to extract significant features following supervised algorithm for differentiating different brain MRI scans to detect and classify presence of brain tumor. As a result, the proposed model achieved 98% accuracy and outperformed the existing methods in the process of diagnosing and classifying brain tumor. The novelty lies in the integration of a deep convolutional feature extractor with a traditional SVM classifier, followed by one of the explainable approach namely Gradient weighted class activation mapping for achieving transparent outcomes based on the two different datasets for enhancing generalization and comparison with other approaches is also done to ensure effectiveness of the proposed model to gain trust of medical experts for speeding up the process of making decisions while diagnosing brain tumor.