Medical Image Quality & AI ReliabilityActive Benchmark / Project 5
License: MITQuantifying Clinical MRI Quality Effects on Deep Learning Performance
Treating Real-World Image Quality Degradation as a Continuous Independent Variable to Define AI Safety Boundaries
Python 3.11+PyTorchMONAISimpleITKOpenCVScikit-ImageResNet-18DenseNet-121EfficientNet-B0
Core Research Ecosystem Vision
Systematically quantifying how continuous objective Image Quality Metrics (SNR, CNR, blur, sharpness, edge strength) affect diagnostic accuracy, calibration error (ECE), and Grad-CAM saliency drift across 225 clinical scans.
Objective Image Quality Metrics (IQMs) & Quality Tiers
Sharpness
Edge definition measured via the variance of Laplacian.
Signal-to-Noise Ratio (SNR)
Signal intensity relative to estimated background noise.
Contrast-to-Noise Ratio (CNR)
Tissue contrast definition relative to image noise.
Entropy
Information content and spatial image complexity.
Blur
Estimated via gradient-based and frequency-domain methods.
Edge Strength
Preservation of anatomical boundaries.
Intensity Uniformity
Assesses bias field and RF coil inhomogeneity.
Resolution Metrics
Pixel spacing, slice thickness, and matrix dimensions.
Quality Stratification TierCohort Characteristics & Degradation
High QualityHigh sharpness, high SNR, minimal noise, clean anatomical boundaries.
Moderate QualityMild blur or noise, clinically clean and acceptable.
Low QualityNoticeable artifacts, visible motion blur, low contrast.
Very Low QualitySignificant degradation, severe motion/blur, but still clinically interpretable.
Execution Commands & Pipeline Workflow
# 1. Train MONAI 3D DenseNet-121 Benchmark:
python src/models/cnn.py --model densenet121 --task task4 --epochs 50 --batch_size 4 --out_dir ./checkpoints/densenet
# 2. Evaluate Degradation Robustness & RRI Metrics:
python src/evaluation/metrics.py --evaluate_robustness --model_dir ./checkpoints/ --out_dir ./results/
# 3. Generate Grad-CAM Explainability Heatmaps under Perturbation:
python src/evaluation/explain.py --model_path ./checkpoints/densenet/best.pth --method gradcam --out_dir ./results/explanations/