How mgi-lab/osocad-ncct-ssfm works
This repository packages a research classifier for opportunistic obstructive coronary artery disease screening on non-contrast chest CT, alongside approved aggregate publication results. The main user workflow is to train or load the classifier, run it on authorized local cohort images and labels, and inspect diagnostic outputs; separate utilities support statistical reanalysis and publication display/export. Patient-level data are not included. The map covers documented capabilities and source-index relationships; unsampled implementations and their detailed wiring are not assumed.
Main entry points: eval_external.py Cohort evaluation · diagnostic_statistics.py Diagnostic statistics · train.py Model training · build_github_english_release.py Public display generation
Components 18
Training
- Slice dataset
code/training/dataset.py - CT classifier
code/training/model.py - Model training
code/training/train.py - Training launcher
code/training/launch.py - Packaged checkpoint
models/best.pth
Inference
- Cohort evaluation
code/inference/eval_external.py - Table 2 inference
code/inference/infer_table2_diagnostic_performance.py
Statistics and exports
- Diagnostic statistics
code/metrics/diagnostic_statistics.py - Checkpoint table export
code/metrics/export_checkpoint_table2_excel.py - Manuscript table export
code/plotting/export_manuscript_tables.py - Calcium figure previews
code/plotting/redraw_calcium_sens_npv_from_excel.py
Publication bundle
- Approved aggregate results
paper_plots - Public display generation
code/plotting/build_github_english_release.py - Bundle verification
code/validation/verify_submission.py - Release packaging
code/validation/package_release.py - README and index
README.md
Other components
- Researcher
- Local study inputs
Connections 23
- Researcher → Training launcher: starts training
README.md - Local study inputs → Model training: provides cohort data
- Model training → Slice dataset: uses dataset
code/training/train.py - Model training → CT classifier: trains model
code/training/train.py - Model training → Packaged checkpoint: saves checkpoint
- Researcher → Cohort evaluation: runs evaluation
README.md - Local study inputs → Cohort evaluation: supplies local cohorts
code/inference/eval_external.py - Cohort evaluation → Slice dataset: uses dataset
code/inference/eval_external.py - Cohort evaluation → CT classifier: loads classifier
code/inference/eval_external.py - Cohort evaluation → Diagnostic statistics: computes metrics
code/inference/eval_external.py - Researcher → Table 2 inference: runs comparison
README.md - Table 2 inference → Diagnostic statistics: uses statistics
code/inference/infer_table2_diagnostic_performance.py - Checkpoint table export → Diagnostic statistics: uses statistics
code/metrics/export_checkpoint_table2_excel.py - Manuscript table export → Diagnostic statistics: uses statistics
code/plotting/export_manuscript_tables.py - Researcher → Manuscript table export: exports tables
README.md - Researcher → Calcium figure previews: creates previews
README.md - Calcium figure previews → Diagnostic statistics: uses statistics
code/plotting/redraw_calcium_sens_npv_from_excel.py - Researcher → Public display generation: checks or refreshes
README.md - Approved aggregate results → Public display generation: renders displays
code/plotting/build_github_english_release.py - Public display generation → README and index: updates pages
code/plotting/build_github_english_release.py - Researcher → Bundle verification: verifies bundle
README.md - Researcher → Release packaging: packages artifacts
code/validation/package_release.py - Release packaging → Bundle verification: validates bundle
code/validation/package_release.py