https://github.com/fitushar/triannot
Tri-Reader: An Open-Access, Multi-Stage AI Pipeline for First-Pass Lung Nodule Annotation in Screening CT
https://github.com/fitushar/triannot
annotation-tool computed-tomography
Last synced: 5 months ago
JSON representation
Tri-Reader: An Open-Access, Multi-Stage AI Pipeline for First-Pass Lung Nodule Annotation in Screening CT
- Host: GitHub
- URL: https://github.com/fitushar/triannot
- Owner: fitushar
- Created: 2025-09-29T19:06:52.000Z (10 months ago)
- Default Branch: main
- Last Pushed: 2026-01-28T08:33:45.000Z (6 months ago)
- Last Synced: 2026-01-28T23:51:10.540Z (6 months ago)
- Topics: annotation-tool, computed-tomography
- Homepage:
- Size: 4.15 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Tri-Reader
**Tri-Reader: An Open-Access, Multi-Stage AI Pipeline for First-Pass Lung Nodule Annotation in Screening CT**
[](https://creativecommons.org/licenses/by-nc/4.0/)
[](https://hub.docker.com/r/ft42/pins)
[](https://python.org)
[](https://simpleitk.org)
[](https://pytorch.org)
[](https://monai.io)
[](https://github.com/fitushar/PiNS)
[](https://github.com/fitushar/CaNA)
Using multiple open-access models trained on public datasets, we developed Tri-Reader, a
comprehensive, freely available pipeline that integrates lung segmentation, nodule detection, and
malignancy classification into a unified tri-stage workflow. The pipeline is designed to
prioritize sensitivity while reducing the candidate burden for annotators. To ensure accuracy and
generalizability across diverse practices, we evaluated Tri-Reader on multiple internal and external
datasets as compared with expert annotations and dataset-provided reference standards
### Citation Manuscript
[](https://arxiv.org/pdf/2601.19380)
```ruby
@misc{tushar2026trireaderopenaccessmultistageai,
title={Tri-Reader: An Open-Access, Multi-Stage AI Pipeline for First-Pass Lung Nodule Annotation in Screening CT},
author={Fakrul Islam Tushar and Joseph Y. Lo},
year={2026},
eprint={2601.19380},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2601.19380},
}
```
**Tri-Reader** tri-stage workflow for first-pass lung nodule annotation in screening CT.
After lung segmentation (VISTA3D) removes extra-pulmonary candidates, Stage 1 performs consensus
CADe using two complementary detection models (CADe-ROIonly and the false-positive–aware
CADe-FPaware trained via strategic hard-negative mining); candidates detected by both are promoted
to confidence = 1.0. Stage 2 applies ensemble malignancy scoring to disagreement candidates using
two CADx classifiers trained on complementary distributions (DLCS24-CADxr50SWS and LUNA25-
CADxr50); candidates with an averaged $CADx ≥ 0.10$ are promoted (confidence = 0.5). Stage 3 retains
remaining candidates meeting $CADe ≥ 0.20$ (confidence = 0.2), yielding a single merged candidate list
with tiered confidence intended to reduce annotator workload while preserving sensitivity; an optional
rule-based NLP module can link report-derived descriptors to candidates when radiology reports are
available.
**Below candidate review panel for SAMPLE_0124 (68-year-old; Intgmultiomics cohort, SAMPLE_0124.nii).**
The leftmost tile is the dataset-provided ground-truth nodule; the remaining tiles are additional candidates proposed by our pipeline. Each tile shows the axial CT slice (left) and a magnified ROI (right); the yellow square marks the candidate location. The header above each tile reports ANod (annotation-consensus score), CADx (malignancy probability), CADe (detector confidence), Lrg.ax (largest axis, mm), lobar location, and slice index. Frame color follows the cancer-risk scale shown in the bottom bar (<0.25, ≥0.25, ≥0.50, ≥0.75).
## 🚀 Updates
- **[1] 09/30/2025** - 📢 Created Github and Huggingface repo
- **[2] 28/01/2025** - 📂 Mnauscript (Pre-print) [](https://arxiv.org/pdf/2601.19380)
- **[3]** - 📂 Public release of **19K NLST CT Annotations** .
- **[4]** - 📂 Public release of **CT RATE Validation Dataset Annotations** .