[2Win5] Poster session 2

[2Win5-35] Toward Automated Orthodontic Diagnosis Using Large Language Models

〇Soichiro Sugihara¹, Tomoyuki Kajiwara¹, Naoki Ikeda², Chihiro Tanikawa², Takashi Ninomiya¹ (1.Ehime University, 2.Osaka University)

Keywords:Orthodontic Treatment, Automated Diagnosis, Large Language Models

We improve the performance of automated orthodontic diagnosis with large language models (LLMs). Previous studies have tackled orthodontic diagnosis from medical findings as a multi-label text classification task employing machine/deep learning techniques. In this study, we improve the performance of this task with LLMs, which have been remarkably successful in recent years. Although we targeted Japanese medical findings, we employed English for our prompts given to the LLM, which improved performance compared to using those in Japanese. Furthermore, we achieved a significant performance improvement with the JSON format of the prompts. Experimental results show that LLM with prompt engineering achieves state-of-the-art performance for orthodontic diagnosis.