ScholarDeck

AI / informaticsmedRxiv : the preprint server for health sciences · 2026

Clinically aligned rationale generation for glaucoma subtype classification via a knowledge-distilled language model.

Mousa Moradi, Asahi Fujita, Niloufar Bineshfar … Nazlee Zebardast

Read paper

OphthalmologyAI / informatics

Fine-tuned AI model achieved high accuracy in generating glaucoma subtype rationales.

Clinically aligned rationale generation for glaucoma subtype classification via a knowledge-distilled language model.

Mousa Moradi et al. · medRxiv : the preprint server for health sciences · 2026

Background

Automated glaucoma subtype classification from clinical notes remains clinically unactionable without subspecialty-aligned explanations supporting clinician-facing deployment.

ROUGE-L 0.792

Results

This score shows how closely the AI's explanations matched expert explanations

n = 2,660 de-identified ophthalmology

More results

We extended our Ci-SSGAN with a GPT-5.2-to-Qwen3-8B teacher-distilled reasoning module, fine-tuning Qwen3-8B on 2,660 de-identified ophthalmology notes using expert-reviewed rationales.

“

Conclusion

On 294 notes, the fine-tuned model achieved ROUGE-L 0.792 ± 0.013 and BERTScore F1 0.955 ± 0.004, surpassing eight zero-shot comparators including GPT-4o and GPT-4.1, establishing privacy-preserving distillation as a path to interpretable AI.

Post

Clinically aligned rationale generation for glaucoma subtype classification via a knowledge-distilled language model.

Fine-tuned AI model achieved high accuracy in generating glaucoma subtype rationales.

Related papers