MRI Brain tissue segmentation using ensemble learning through dl

This project develops a deep learning-based system for multi-class segmentation of brain tissues from MRI scans. The primary objective is to achieve precise pixel-level classification of brain structures using advanced convolutional neural network architectures. The segmentation task classifies pixels into seven distinct classes: Background, Air cavities, White matter, Gray matter, Cerebrospinal fluid (CSF), Bone, and Non-brain soft tissue, with ground-truth masks provided in colored format. Two state-of-the-art models—standard U-Net and Attention U-Net—are implemented and trained separately on 2D MRI slices derived from 3D volumes. Model performance is thoroughly evaluated using Intersection over Union (IoU) and accuracy metrics to measure effectiveness in tissue delineation. Furthermore, a user-friendly local web application is developed, incorporating user registration and login features. Users can upload MRI slices and obtain segmented outputs with color-coded tissue labels. This work advances deep learning applications in brain tissue segmentation and delivers a practical tool for clinical and research use.

Keywords:

MRI brain tissue segmentation, deep learning, U-Net, Attention U-Net, multi-class segmentation, 3D to 2D preprocessing, IoU evaluation, accuracy metrics, web application, local deployment