RetiJev: Fast and Source-Grounded Ophthalmic Decision Models

Bo Liu; Lianyu Wang; Meng Wang; Ching-Yu Cheng; Huazhu Fu

cs.CLOct 22, 2026

RetiJev: Fast and Source-Grounded Ophthalmic Decision Models

Authors: Bo Liu, Lianyu Wang, Meng Wang, Ching-Yu Cheng, Huazhu Fu

Organizations: Centre for Innovation and Precision Eye Health, Yong Loo Lin School of Medicine, National University of Singapore, Singapore 119228, Singapore. · Department of Ophthalmology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore 119228, Singapore. · University of Washington, Seattle, WA, USA. · Singapore Eye Research Institute, Singapore National Eye Centre, Singapore 169856, Singapore. ·

Abstract

Ophthalmic clinical decision support requires fast, accurate, and privacy-preserving inference, yet autoregressive large language models (LLMs) incur substantial latency and computational overhead for decisions among predefined actions. We therefore investigate a decision-oriented alternative to generative reasoning. Specifically, we first introduce RetiJev-Data, a source-grounded ophthalmology corpus comprising 78,158 clinical-decision and diagnosis questions derived from articles in PubMed Central and a MeSH vocabulary covering 531 eye diseases, and 50,074 guideline-grading questions spanning 11 grading systems, along with a 24,662-question evaluation benchmark. Based on this, we develop RetiJev, a family of 0.8B-9B decision models that directly score candidate answers without autoregressive decoding. Across all model sizes, RetiJev outperforms both a 27B LLM and the commercial Jev 1.13 decision model on guideline grading by 10.4-23.0 points. RetiJev-9B further achieves the highest accuracy on clinical decisions, improving over both baselines by 5.4-6.5 points while matching them on diagnosis. It answers a case in 158 ms on one GPU, 3.5x faster than LLM generation and 470x faster than reasoning. Models are released at https://github.com/Awenbocc/RetiJev.

Figures & tables

Panel of Figure 1 — Figure 1: Inference latency and peak GPU memory for ophthalmic case-level decision making. Latency is measured as the median per-case inference time on a single NVIDIA H200 GPU with batch size 1 (averaging 2.65 questions per case). All three RetiJev variants achieve 0.13–0.16 s latency with 3.3–31.5 GB of peak GPU memory, while RetiJev-0.8B also supports CPU-only inference on eight cores. The 27B LLM requires approximately 51 GB of GPU memory and exceeds one minute per case with reasoning enabled.

Panel of Figure 2 — Figure 2: Construction pipeline of RetiJev-Data. Clinical decisions (left): ophthalmic literature is retrieved and organized into evidence bundles. Conditional rules guide case synthesis, followed by evidence filtering and independent label verification. Guideline grading (right): labels are computed from structured case parameters, and independent parameter re-extraction verifies the fidelity of generated clinical notes.

Training	94,561	Development	9,009
Main test sets (16,332 questions)
Decision-Unseen	4,625	Decision-Seen	3,106
Diagnosis-Unseen	164	Diagnosis-Seen	224
Grading-OOD	3,983	Grading-Hard	4,230
Grading reference sets (8,330 questions)
Grading-Standard	4,167	Grading-Paraphrase	4,163

Table 1: Training data and evaluation sets. Counts denote the number of questions. Training data composition and evaluation set definitions are detailed in Tables 14 and 8 .

Figure 3: RetiJev architecture and training objective. (a) Independent state–question sequences pass through a shared backbone with frozen base weights and trainable LoRA adapters on attention, DeltaNet and MLP projections. (b) A shared pointer head projects the hidden state at $\langle\textsc{d}\rangle$ into a query to score the candidate representations. (c) Hard labels supervise all questions, with an additional ordinal loss for ordered ratings. Clinical-decision questions can also receive offline teacher distributions and paraphrase-consistency targets when available. Neither auxiliary target is used for guideline grading. Only the adapters and pointer head are trained.

Figure 4: Shared-context inference. (a) A generative LLM processes the patient state separately for each question and autoregressively decodes an answer. (b) RetiJev encodes the patient state once, replicates the cached attention and recurrent states across independent question branches, and scores all branches in parallel without autoregressive decoding.

Ophthalmology

See all 41 papers

Guideline-grounded retrieval-augmented generation for ophthalmic clinical decision support
Mar 23, 2026Shuying Chen, Sen Cui, Zhong CaoMultimodal RAG Retrieval-Augmented Generation
School of Artificial Intelligence and Data Science University of International Business and Economics Beijing, China · Tsinghua University Beijing, China · Heidelberg Institute of Global Health, Faculty of Medicine and University Hospital Heidelberg University Heidelberg, Germany
Decision language models as affordable and rapid alternatives to LLM judges of clinical generative AI
Oct 20, 2026Rotem Gershon, Seffi Cohen, Dean Ariel +11LLM-as-a-Judge Clinical Language Model Evaluation
Innovation Division, Clalit Research Institute, Clalit Health Services, Ramat-Gan, Israel. · Stein Faculty of Computer and Information Science, Ben Gurion University of the Negev, Be’er Sheva, Israel. · Department of Family Medicine, Clalit Health Services, Tel Aviv, Israel. +4
VOLMO: Versatile and Open Large Models for Ophthalmology
Mar 25, 2026Zhenyue Qin, Younjoon Chung, Elijah Lee +16Medical VLMs Medical LLMs
Department of Biomedical Informatics & Data Science, Yale University · Ray and Stephanie Lane Computational Biology Department, Carnegie Mellon University · Yong Loo Lin School of Medicine, National University of Singapore +4

Mar 23, 2026J/K move · Enter open · S save

Guideline-grounded retrieval-augmented generation for ophthalmic clinical decision support

Shuying Chen, Sen Cui, Zhong Cao

School of Artificial Intelligence and Data Science University of International Business and Economics Beijing, China · Tsinghua University Beijing, China · Heidelberg Institute of Global Health, Faculty of Medicine and University Hospital Heidelberg University Heidelberg, Germany

Multimodal RAG Retrieval-Augmented Generation Healthcare Medical Diagnosis Clinical Decision Support Ophthalmology+1

Open paper AskPDF arXiv

In this work, we propose Oph-Guid-RAG, a multimodal visual RAG system for ophthalmology clinical question answering and decision support. We treat each guideline page as an independent evidence unit and directly retrieve page images, preserving tables, flowcharts, and layout information. We further design a controllable retrieval framework with routing and filtering, which selectively introduces external evidence and reduces noise. The system integrates query decomposition, query rewriting, retrieval, reranking, and multimodal reasoning, and provides traceable outputs with guideline page references. We evaluate our method on HealthBench using a doctor-based scoring protocol. On the hard subset, our approach improves the overall score from 0.2969 to 0.3861 (+0.0892, +30.0%) compared to GPT-5.2, and achieves higher accuracy, improving from 0.5956 to 0.6576 (+0.0620, +10.4%). Compared to GPT-5.4, our method achieves a larger accuracy gain of +0.1289 (+24.4%). These results show that our method is more effective on challenging cases that require precise, evidence-based reasoning. Ablation studies further show that reranking, routing, and retrieval design are critical for stable performance, especially under difficult settings. Overall, we show how combining visionbased retrieval with controllable reasoning can improve evidence grounding and robustness in clinical AI applications,while pointing out that further work is needed to be more complete.

Abstract

Ophthalmology

Explore similar work

Guideline-grounded retrieval-augmented generation for ophthalmic clinical decision support

Decision language models as affordable and rapid alternatives to LLM judges of clinical generative AI

VOLMO: Versatile and Open Large Models for Ophthalmology