Project Description
I’m putting together a deep-learning workflow that can read MRI scans and immediately show why it reached each conclusion. The core of the job is to train a high-performing model on my MRI dataset and then weave in SHAP for feature-level interpretation and Grad-CAM for pixel-level heat-maps.
The system needs to surface three kinds of insight for every study that runs through it:
• a clear diagnosis justification that highlights the decisive regions of each image,
• tailored treatment recommendations drawn from the predicted class probabilities, and
• patient-specific risk factors ranked by SHAP values.
I already have raw MRI images and basic labels; you would handle the full pipeline—from preprocessing and augmentation through model tuning, inference, and explanation generation. Code should be delivered in clean, modular Python (PyTorch preferred, but TensorFlow is fine if you make a strong case) with Jupyter notebooks for reproducibility.
Acceptance criteria
• Model must reach the benchmark accuracy we agree on during kickoff.
• Grad-CAM overlays and SHAP plots render automatically for each study.
• Outputs exportable as JSON for structured data and PNG for visuals.
• README explains setup, training steps, and how to plug in new MRI data.
If you have prior work combining SHAP and CAM techniques—especially in medical imaging—let’s talk.