Enhanced Brain Tumor Segmentation and Classification using DeepUNet3+ and Convolutional Boosting

Brain tumor is a deadly disease and detection of brain tumors poses a significant diagnostic challenge for radiologists due to the diverse nature of tumor cells. To address this challenge and enhance diagnostic accuracy, computer-aided diagnosis systems have emerged as promising tools, particularly in the context of magnetic resonance imaging (MRI). However, current applications employing pre-trained models face difficulties in extracting features and predict its type from medical images, which differ significantly from natural images. To overcome this issue, this paper proposes an efficient framework for brain tumor prediction combining image segmentation, CNN and Ada Boosting. For image segmentation we have introduced DeepUNet3+, for prediction the type of tumor we used a boosted fusion model which is combination of CNN and Ada Boost. We take 2870 brain tumor images, which are categorized into glioma, meningioma, pituitary gland tumors, and healthy brain samples. This approach integrates DeepUNet3+ for segmentation, achieving noteworthy results with an F1 score of 0.93, recall value of 0.92, and precision value of 0.92. To further improve diagnostic capabilities, we utilize a combination of Convolutional Neural Network (CNN) and boosting models, resulting in an impressive prediction accuracy of 96.03%. The Boosted FusionNet leverages the strengths of DeepUNet3+, --CNN, and boosting techniques, providing a comprehensive solution for the intricate task of brain tumor segmentation and detection. This innovative approach holds significant promise for enhancing the efficiency and reliability of early diagnosis, ultimately contributing to more effective and timely medical interventions.