Research map / Large language models / Retrieval and knowledge
Evidential Reasoning Advances Interpretable Real-World Disease Screening
Chenyu Lian, Hong-Yu Zhou, Jing Qin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add evidence-aware reasoning on top of evidence retrieval (ablation) | EviScreen · JSIEC | Spe@95%R: 91.48 -> 94.74 | Improved |
| add evidence retrieval on top of evidence-aware reasoning (ablation) | EviScreen · RIADD | Spe@95%R: 61.86 -> 72.92 | Improved |
| add evidence-aware reasoning to retrieval only (ablation) | EviScreen · RIADD | AUROC: 90.42 -> 91.32 | Improved |
| remove evidence retrieval (reasoning only) vs retrieval only (ablation) | EviScreen · RIADD | AUROC: 90.42 -> 86.12 | Hurt |
| use dual knowledge banks with contrastive retrieval instead of normal-only PatchCore (same foundation models) | Training-free variant · JSIEC | AUROC: 94.96 -> 96.76 | Improved |
| replace fixed learned prototypes (CIPL) with scalable dual knowledge banks | EviScreen · CheXpert | Spe@95%R: 68.09 -> 84.04 | Improved |
| evidential reasoning vs direct fine-tuning of foundation model (FM) | EviScreen · JSIEC | Spe@100%R: 78.39 -> 91.27 | Improved |
| evidential reasoning vs direct fine-tuning of foundation model (FM) | EviScreen · CheXpert | CSR: 57.41 -> 70.59 | Improved |
| increase number of samples in dual knowledge banks (ablation) | Training-free variant · ophthalmology validation | Spe@X%R | Improved |
| use foundation-model features in PatchCore instead of ImageNet-pretrained (ablation) | PatchCore · four test sets | screening metrics: 92.12 -> 94.96 AUROC (JSIEC) | Improved |
Closest papers by meaning
- RAD: Towards Trustworthy Retrieval-Augmented Multi-modal Clinical Diagnosis (NeurIPS 2025 · similarity 0.85)
- DDx-TRACE: A Benchmark for Medical Diagnostic Trajectories in VLMs (NeurIPS 2026 · similarity 0.84)
- Medical thinking with multiple images (ICLR 2026 · similarity 0.84)
- M3CoTBench: Benchmark Chain-of-Thought of MLLMs in Medical Image Understanding (ICLR 2026 · similarity 0.84)
- MedHorizon: Towards Long-context Medical Video Understanding in the Wild (NeurIPS 2026 · similarity 0.83)
- EMAD: Evidence-Centric Grounded Multimodal Diagnosis for Alzheimer's Disease (CVPR 2026 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
