Brain Tumor Segmentation and Classification using Deep neural networks

Accurate segmentation and recognition of brain tumors are vital in medical research for correct diagnosis and strategic treatment planning. This project proposes an innovative approach by combining segmentation and classification to expand the accuracy and efficacy of brain tumor evaluation. At first, a UNet model is implemented to demarcate the brain tumors in 2D MRI images. This phase involves isolating the tumor from the background, resulting in segmented images. Subsequently, the original MRI images are concatenated with the segmented images to create enhanced inputs. Following the segmentation, a CNN model is developed to classify brain tumor types. The concatenated images are inputted into the CNN model for classification, enabling the network to learn from both the original MRI images and their segmented versions. Validations are carried out on a dataset comprising 2D MRI slices of brain tumors, including glioma, meningioma, and pituitary tumor types. The proposed approach yielded favorable outcomes in both segmentation and classification accuracy. The segmentation phase attained a Dice similarity coefficient (DSC) of 76% for tumor regions, while the classification model achieved an accuracy of 92% on the test dataset.