suitch's picture
Point to thesis repository
4157388 verified
|
Raw History Blame Contribute Delete
2.43 kB
---
license: cc-by-nc-sa-4.0
base_model: Qwen/Qwen3.5-4B
library_name: peft
pipeline_tag: image-text-to-text
tags:
- medical
- radiology
- ct
- report-generation
- ct-rate
- lora
---
# Baseline report decoder
Part of the master's thesis *From Scan to Report: Detecting, Segmenting and Reporting Findings in Chest CT* (code: https://github.com/Switchharsh/Switchharsh-From_Scan_to_Report_Master_Thesis; thesis: https://github.com/Switchharsh/Switchharsh-From_Scan_to_Report_Master_Thesis/tree/main/thesis).
Baseline scan-to-report decoder: FlexiCT tokens -> projector -> Qwen3.5-4B + LoRA, trained 750 steps on 12k CT-RATE training reports (lr 1e-4, batch 2 x 8 accumulation, seed 0, plain cross-entropy).
| condition | CRG | macro-F1 |
|---|---|---|
| real scan | 0.3499 | 0.0854 |
| another patient's scan | 0.3492 | 0.0457 |
Three-seed mean (seeds 0-2): CRG 0.3567 +- 0.0062 (real) vs 0.3534 +- 0.0040 (mismatched). The baseline is barely scan-dependent on CRG; it mostly writes a near-normal template (1.1 findings per report).
## How it fits together
- Input: the 8x8x8 FlexiCT token grid of a chest CT (`flexict/tokens8`, 864-dim per token) from CT-RATE.
- `projector.pt`: MLP (864 -> 2560 -> 2560, LayerNorm) that maps the visual tokens into the LLM embedding space.
- `adapter_model.safetensors` + `adapter_config.json`: LoRA (r=16, alpha=32) on [Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B).
- Loading and generation code (model wrapper, data, evaluation): https://github.com/Switchharsh/Switchharsh-From_Scan_to_Report_Master_Thesis (`ctbench/rgen/model.py`, `evaluate.py`). The adapter is not a stand-alone chat model: the projector and the FlexiCT tokens are required.
- Task instruction used in every run: `Write the findings and impression for this chest CT.` Greedy decoding, at most 512 new tokens.
## Evaluation protocol
1,000 patient-sampled CT-RATE validation cases; CRG (micro) and macro-F1 with the RadBERT label reader (`suitch/radbert-english-ctrate-classifier`, threshold 0.5). Control = another patient's scan tokens (mismatched). Numbers are from the thesis *From Scan to Report: Detecting, Segmenting and Reporting Findings in Chest CT*.
## Limitations / licence
Research use only, not for clinical use. Trained on CT-RATE (CC BY-NC-SA 4.0), so the weights are released under the same non-commercial terms. Reports can contain false findings; see the thesis for the error analysis.