Multimodal Retrieval-Augmented Generation on IU-Chest
68.31AccuracyStandard RAG + BAIR
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Standard RAG + BAIRModel=MedGemma-4B2026.05 | 68.31 | 24.25 | 8.4 | 2.89 | 18.38 | 43.75 | 7.93 | |
| MAD-RAG + BAIRModel=MedGemma-4B2026.05 | 68.25 | 24.19 | 8.45 | 2.86 | 18.77 | 44.02 | 8.22 | |
| MAD-RAGModel=MedGemma-4B2026.05 | 68.03 | 24.58 | 9.01 | 2.73 | 17.64 | 40.79 | 7.88 | |
| MS-PoE + BAIRModel=MedGemma-4B2026.05 | 67.6 | 21.01 | 7.43 | 2.83 | 20.33 | 50.74 | 7.79 | |
| MS-PoEModel=MedGemma-4B2026.05 | 67.17 | 20.68 | 7.85 | 2.63 | 18.99 | 46.7 | 7.42 | |
| MS-PoE + BAIRModel=CheXagent-8B2026.05 | 66.6 | 11.72 | 5.05 | 2.32 | 13.95 | 47.78 | 3.19 | |
| MS-PoEModel=CheXagent-8B2026.05 | 66.45 | 14.19 | 6.62 | 2.14 | 12.96 | 40.3 | 3.87 | |
| Standard RAGModel=MedGemma-4B2026.05 | 66.33 | 28 | 13.56 | 2.06 | — | — | — | |
| LongLLMLingua + BAIRModel=CheXagent-8B2026.05 | 66.24 | 16.64 | 8.36 | 1.99 | 19.16 | 47.96 | 9.32 | |
| Standard RAG + BAIRModel=CheXagent-8B2026.05 | 66.17 | 12.38 | 6.04 | 2.05 | 10.47 | 39.35 | 1.56 | |
| MAD-RAG + BAIRModel=CheXagent-8B2026.05 | 66.08 | 11.83 | 5.87 | 2.02 | 11.73 | 42.35 | 2.24 | |
| LongLLMLingua + BAIRModel=MedGemma-4B2026.05 | 65.91 | 25.21 | 12.46 | 2.02 | 21.74 | 43.41 | 12.14 | |
| Visual-focus Instr. + BAIRModel=MedGemma-4B2026.05 | 65.69 | 24.48 | 12.33 | 1.99 | 16.37 | 31.59 | 9.43 | |
| LongLLMLinguaModel=MedGemma-4B2026.05 | 65.66 | 25.14 | 12.79 | 1.97 | 21.73 | 42.41 | 12.33 | |
| Visual-focus Instr. + BAIRModel=CheXagent-8B2026.05 | 65.58 | 12.89 | 7.19 | 1.79 | 7.95 | 28.03 | 1.5 | |
| BaselineModel=CheXagent-8B2026.05 | 65.51 | — | — | — | — | — | — | |
| Visual-focus Instr.Model=MedGemma-4B2026.05 | 65.38 | 26.68 | 14.13 | 1.89 | 16.74 | 30.5 | 10.35 | |
| LongLLMLinguaModel=CheXagent-8B2026.05 | 65.33 | 23.63 | 13.63 | 1.73 | 21.85 | 38.43 | 15.13 | |
| Visual-focus Instr.Model=CheXagent-8B2026.05 | 64.94 | 15.89 | 9.81 | 1.62 | 4.04 | 10.81 | 1.87 | |
| MAD-RAGModel=CheXagent-8B2026.05 | 64.79 | 16.16 | 10.19 | 1.59 | 1.18 | 2.93 | 0.57 | |
| Standard RAGModel=CheXagent-8B2026.05 | 64.56 | 15.88 | 10.36 | 1.53 | — | — | — | |
| BaselineModel=MedGemma-4B2026.05 | 63.76 | — | — | — | — | — | — |