How mwasifanwar/medvision-ai works
Project Type and Purpose
• This is a modular, pipeline-oriented medical imaging diagnostic assistant written in Python.
• Purpose: Ingest DICOM images, preprocess and normalize them, run modality-specific AI analyzers, fuse multi-modal evidence, correlate findings with clinical context, and generate structured reports.
File Structure Analysis
• main.py
– Application entry point (likely a Streamlit app or CLI orchestrator).
• core/
– dicom_processor.py: DICOM loading, decoding, metadata extraction, normalization.
– model_manager.py: Downloading, versioning, and loading pre-trained models.
– predictor.py: Dispatches to modality-specific AI analyzers and fusion engine.
– clinical_correlator.py: Combines image findings with patient context for risk stratification.
– report_generator.py: Summarizes findings, grades severity, produces recommendations.
• utils/
– config.py: Parameter definitions (thresholds, file paths).
– medical_metrics.py: Dice loss, uncertainty quantification, other domain metrics.
– visualization.py: Plotting, overlay of masks, confidence heatmaps.
• requirements.txt: Core dependencies (PyTorch, MONAI, SimpleITK, Streamlit).
Insights from README
• Multi-stage diagnostic pipeline with stages: DICOM Processor → Preprocessor → X-Ray/MRI/CT Analyzers → Fusion Engine → Clinical Correlator → Report Generator → Quality Assurance.
• Key technologies: PyTorch + MONAI for models, SimpleITK/OpenCV for image ops, Streamlit for UI, Docker for deployment, PACS integration for DICOM storage.