Manya-ga/Cerebra

Voxel-Level Brain Tumor Segmentation Using 3D Deep Learning Cerebra Voxel-Level Brain Tumor Segmentation Using 3D Deep Learning Cerebra is a deep learning system for automated segmentation of brain tumors from multi-modal 3D MRI scans. It identifies and delineates distinct tumor sub-regions at the voxel level, with the goal of supporting radiologists in faster, more consistent diagnosis and treatment planning. Built on a 3D U-Net architecture trained on the BraTS dataset, Cerebra combines a MONAI-based inference pipeline with a lightweight web interface for uploading scans and reviewing segmentation results. Overview Manual segmentation of brain tumors from MRI is time-consuming and subject to inter-observer variability. Cerebra automates this process by processing four MRI modalities in parallel — T1, T1CE, T2, and FLAIR — and producing a segmentation mask that separates tumor tissue into clinically relevant sub-regions. Core capabilities: Automated 3D segmentation of brain tumors from multi-modal MRI input Voxel-level prediction across three tumor sub-regions (whole tumor, tumor core, enhancing tumor) Web-based upload and inference pipeline with result visualization Model built on a 3D U-Net architecture using the MONAI framework Tech Stack LayerTechnologiesFrontendReact, Vite, Tailwind CSSBackendFastAPI, PythonDeep LearningPyTorch, MONAIImaging & PreprocessingNiBabel, SimpleITK, NumPy, OpenCVVisualizationMatplotlib, Plotly Dataset Cerebra is trained on the BraTS (Brain Tumor Segmentation Challenge) dataset, which provides skull-stripped, co-registered MRI volumes across four modalities: T1 — native T1CE — contrast-enhanced T2 — T2-weighted FLAIR — fluid-attenuated inversion recovery Ground-truth annotations label three tumor tissue classes: NCR — Necrotic core ED — Peritumoral edema ET — Enhancing tumor These are combined at inference time into the three clinically standard evaluation regions: Whole Tumor (WT), Tumor Core (TC), and Enhancing Tumor (ET). Project Structure Cerebra/ ├── frontend/ …