NeuroSpine AI

A Clinical-Grade Deep Learning System for Spine X-Ray Diagnosis & Reporting

Overview

NeuroSpine AI is an advanced medical imaging web application designed to assist radiologists in the detection and classification of spinal abnormalities. It utilizes an Ensemble Deep Learning Model combining ConvNeXt Tiny and EfficientNet-B3 to achieve state-of-the-art accuracy on lumbar spine X-rays.

Beyond simple classification, the system provides Explainable AI (XAI) visuals using Grad-CAM++ to generate high-resolution heatmaps that pinpoint the exact location of injuries. It also features an automated Radiology Report Generator that creates professional PDF reports with clinical recommendations.

Key Features

• ** High-Accuracy Ensemble:** Uses soft-voting ensemble learning with Test-Time Augmentation (TTA) to combine predictions from ConvNeXt and EfficientNet architectures.
• ** Clinical Explainability:** Generates high-resolution Grad-CAM++ heatmaps overlaid on the original high-res X-ray to visually verify the model's focus.
• ** Advanced Preprocessing:** Implements CLAHE (Contrast Limited Adaptive Histogram Equalization) to enhance bone trabecular structure and edge visibility in low-contrast scans.
• ** Automated Reporting:** Instantly generates a downloadable PDF Radiology Report following RSNA standards, complete with patient info, findings, impression, graphs, and heatmaps.
• ** Professional Interface:** A "Cockpit-style" single-page dashboard designed for efficiency, featuring a drag-and-drop diagnostic workflow.

Tech Stack

• Deep Learning: PyTorch, Torchvision (ConvNeXt, EfficientNet)
• Backend: Flask (Python)
• Image Processing: OpenCV, NumPy, Pillow, Matplotlib
• Frontend: HTML5, CSS3 (Grid Layout), JavaScript (ES6)
• Reporting: jsPDF, Chart.js logic

Project Structure

NeuroSpine-AI/
├── app.py                 # Main Flask Application & Model Inference Logic
├── best_model_convnext.pth # Trained Weights for ConvNeXt Model
├── best_model_effnet.pth   # Trained Weights for EfficientNet Model
├── static/
│   ├── styles.css         # Professional Dashboard Styling
│   └── script.js          # Frontend Logic & PDF Report Generation
├── templates/
│   └── index.html         # Main Web Interface
└── README.md              # Documentation