Skip to content

Commit bb039e1

Browse files
committed
readme updated
1 parent 865ff7b commit bb039e1

1 file changed

Lines changed: 14 additions & 4 deletions

File tree

README.md

Lines changed: 14 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -4,9 +4,9 @@ BSc thesis project. The question: does adding structured ED triage data (vitals,
44

55
Short answer: calibration improves, discrimination does not.
66

7-
**Data:** MIMIC-CXR-JPG v2.1.0 + MIMIC-IV-ED (PhysioNet credentialed access required)
7+
**Data:** MIMIC-CXR-JPG v2.1.0 + MIMIC-IV v2.2 + MIMIC-IV-ED v2.2 (PhysioNet credentialed access required)
88
**Cohort:** 9,137 ED-anchored studies, 80/10/10 patient-level temporal split, test set n = 1,075
9-
**Backbone:** DenseNet-121 pretrained on 14-label CheXpert task, fine-tuned for binary pneumonia
9+
**Backbone:** DenseNet-121 pretrained on the multilabel CheXpert-style task using 41,214 non-ED MIMIC-CXR-JPG studies, then fine-tuned for binary pneumonia
1010

1111
---
1212

@@ -116,7 +116,7 @@ docker run --gpus all \
116116
multimodal-pneumonia make all
117117
```
118118

119-
Trained weights (~800 MB–1.2 GB per checkpoint) are not in the repository. Contact the author for access.
119+
Trained weights (~800 MB–1.2 GB per checkpoint) are not redistributed under the PhysioNet Data Use Agreement. To obtain weights, complete PhysioNet credentialing and re-run `make all` after configuring `configs/paths.local.yaml`.
120120

121121
---
122122

@@ -130,7 +130,17 @@ python -m pytest tests/ -v
130130

131131
---
132132

133+
## Dashboard
134+
135+
```bash
136+
streamlit run streamlit_app.py
137+
```
138+
139+
The dashboard reads pre-computed metrics from `artifacts/evaluation/` and runs without MIMIC access for browsing results, calibration plots, bootstrap comparisons, and Grad-CAM overlays.
140+
141+
---
142+
133143
## Data compliance
134144

135-
Uses MIMIC-CXR-JPG v2.1.0, MIMIC-IV, and MIMIC-IV-ED under the PhysioNet Credentialed Health Data License. No patient-level data is committed to this repository. Reproducing results requires completing PhysioNet credentialing at [physionet.org](https://physionet.org).
145+
Uses MIMIC-CXR-JPG v2.1.0, MIMIC-IV v2.2, and MIMIC-IV-ED v2.2 under the PhysioNet Credentialed Health Data License. No patient-level data is committed to this repository. Reproducing results requires completing PhysioNet credentialing at [physionet.org](https://physionet.org).
136146

0 commit comments

Comments
 (0)