How sjyzy/csmd works
MMCSD is a cervical-spondylosis imaging dataset accompanied by code for evaluating pain classification from medical images. The documented workflow supplies an image modality, trained weights, and clinical labels to roc.py, which selects patient samples, evaluates predictions, computes classification metrics, and saves an ROC plot. The repository tree does not contain the documented image folders; model and checkpoint wiring beyond the sampled excerpts is therefore represented only where documented. image_dataset.py separately implements optional paired-image loading and transforms.
Main entry points: roc.py Model evaluation · image_dataset.py Paired image loader · high_pain_text.xlsx Clinical labels · train_id4.txt Patient splits
Components 9
Dataset inputs
- MRI and CT volumes
- Clinical labels
high_pain_text.xlsx - Patient splits
train_id4.txt - Model weights
assets/best_auc.pth
Evaluation runtime
- Model evaluation
roc.py - Paired image loader
image_dataset.py
Evaluation results
- Classification metrics
roc.py - ROC plot
roc.py
Other components
- Researcher
Connections 9
- Researcher → Model evaluation: runs
- MRI and CT volumes → Model evaluation: provides volumes
- Clinical labels → Model evaluation: provides labels
- Patient splits → Model evaluation: provides IDs
- Model weights → Model evaluation: provides checkpoint
- Model evaluation → Classification metrics: computes
- Model evaluation → ROC plot: saves
- Researcher → Paired image loader: uses
- MRI and CT volumes → Paired image loader: loads pairs