Brain Tumor Detection and Localization: An Inception V3 - Based Classification Followed By RESUNET-Based Segmentation Approach

Abstract

Adults and children alike are at risk from brain tumors. Accurate and prompt detection, on the other hand, can save lives. This research focuses on the identification and localization of brain tumors. Many research has been available on the analysis and classification of brain tumors, but only a few have addressed the issue of feature engineering. To address the difficulties of manual diagnostics and traditional feature-engineering procedures, new methods are required. To reliably segment and identify brain tumors, an automated diagnostic method is required. While progress is being made, automated brain tumor diagnosis still confront hurdles such as low accuracy and a high rate of false-positive outcomes. Deep learning is used to analyse brain tumors in the model described in this work, which improves classification and segmentation. Using Inception-V3 and RESUNET, deep learning is pragmatic for tumor classification and segmentation. On the Inception V3 model, add one extra layer as a head for classifying. The outcomes of these procedures are compared to those of existing methods. The test accuracy of the Inception-V3 with extra classification layer model is 0.9996, while the loss value is 0.0025. The model tversky value for localization and detection is 0.9688, while the model accuracy is 0.9700.