notesum.ai
Published at November 15JRadiEvo: A Japanese Radiology Report Generation Model Enhanced by Evolutionary Optimization of Model Merging
cs.CV
cs.AI
cs.CL
cs.NE
Released Date: November 15, 2024
Authors: Kaito Baba1, Ryota Yagi1, Junichiro Takahashi1, Risa Kishikawa1, Satoshi Kodera1
Aff.: 1Department of Cardiovascular Medicine, The University of Tokyo Hospital, Tokyo, Japan
![[Uncaptioned image]](https://arxiv.org/html/2411.09933v1/extracted/6001278/fig/ex1.jpg)
| BLEU-1 | BLEU-2 | BLEU-3 | BLEU-4 | ROUGE-L | METEOR | |
|---|---|---|---|---|---|---|
| JRadiEvo (ours) | 0.376 | 0.301 | 0.241 | 0.179 | 0.212 | 0.191 |
| LLM instruction-tuned | 0.342 | 0.241 | 0.185 | 0.133 | 0.143 | 0.142 |
| LLM + projector instruction-tuned | 0.345 | 0.245 | 0.189 | 0.136 | 0.147 | 0.146 |
| CheXagent [30] | 0.221 | 0.181 | 0.149 | 0.119 | 0.199 | 0.159 |
| GPT-4o [54] | 0.432 | 0.333 | 0.257 | 0.184 | 0.176 | 0.188 |