Automated MRI Lesion Segmentation Using Multi-Model Deep Learning Techniques

The segmentation of multiple sclerosis lesions in MRI images is crucial for analyzing brain abnormalities and monitoring disease progression. This project proposes a multi-pathway 3D Convolutional Neural Network (3D CNN) combined with Conditional Random Field (CRF) for automated and precise segmentation of brain lesions. The approach integrates several advanced deep learning models, including UNet, UNet++, and DeepLabv3, to extract and process spatial and contextual information efficiently. The use of 3D CNN allows capturing volumetric features of MRI images, while CRF refines the segmented output for better boundary accuracy. The project employs a publicly available brain tumor segmentation dataset to simulate lesion segmentation, enabling experimentation with different architectures and comparison of their performances. A web-based interface using Flask framework is developed for easy interaction, allowing users to upload images and visualize segmentation results. This system also includes modules such as Home, Register, Login, Segmentation, and Logout to ensure structured access and operation. Experimental results demonstrate that the proposed framework improves segmentation accuracy and reduces false-positive regions. This study contributes to automated medical image analysis by combining deep learning with probabilistic graphical models for enhanced lesion detection.

Keywords: Multiple Sclerosis, MRI, Segmentation, 3D CNN, UNet, UNet++, DeepLabv3, Conditional Random Field, Flask