https://github.com/apartsinprojects/cxr-selective-rejection
Selective multipathology chest X-ray classification via rejection — revised R3 manuscript + point-by-point response (highlighted edits)
https://github.com/apartsinprojects/cxr-selective-rejection
Last synced: 11 days ago
JSON representation
Selective multipathology chest X-ray classification via rejection — revised R3 manuscript + point-by-point response (highlighted edits)
- Host: GitHub
- URL: https://github.com/apartsinprojects/cxr-selective-rejection
- Owner: ApartsinProjects
- Created: 2026-06-02T15:52:36.000Z (12 days ago)
- Default Branch: main
- Last Pushed: 2026-06-02T19:55:48.000Z (12 days ago)
- Last Synced: 2026-06-02T21:28:43.432Z (12 days ago)
- Language: HTML
- Size: 14 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Selective Multipathology Chest X-Ray Classification via Rejection Mechanisms (R3)
**Live paper (GitHub Pages): **
Revised manuscript and reviewer response for the Round-3 revision.
- **[index.html](https://apartsinprojects.github.io/cxr-selective-rejection/)**: the revised manuscript. Tracked-edit highlighting versus the Round-2 version uses green for newly added content and yellow for modified content (prose, captions, and table cells). It carries 7 figures and 4 tables, evaluated on three public datasets (NIH ChestX-ray14 n=11,212, MIMIC-CXR n=8,185, PadChest n=5,000) with bootstrap confidence intervals.
- **[response.html](https://apartsinprojects.github.io/cxr-selective-rejection/response.html)**: the point-by-point response to Reviewer 2's three concerns, each with a "Where in the manuscript" pointer to the relevant section, table, figure, or equation, and a Harvard-style reference list.
- **[manuscript.docx](https://apartsinprojects.github.io/cxr-selective-rejection/manuscript.docx)** / **[manuscript_2col.docx](https://apartsinprojects.github.io/cxr-selective-rejection/manuscript_2col.docx)**: single-column and two-column Word versions with the same green/yellow tracked-edit highlighting and native (OMML) equations, suitable for track-changes review.
Headline: at a calibrated operating point, per-pathology entropy-based abstention improves balanced accuracy for all four target pathologies on NIH (gains 0.04 to 0.09 at 0.70 coverage, 95% CIs excluding zero), and the benefit replicates on MIMIC-CXR, on PadChest, on an independently trained model (including cross-dataset on MIMIC), and under leave-one-dataset-out domain shift.
Research code and reproduction pipeline: .