The British journal of ophthalmology · 2026
MiKid-QA: construction and performance evaluation of a paediatric myopia education question-answering model.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Wang Zixun, Yu Jingtao, Zhang Tingyu, Lu Daqian, Li Yifan, Hu Yabin, Zhou Jun, Huang Liqin, Zhang Xiaoling, Hu Xiaoxue, Xie Xueshuo, Yan Xu, Yang Jing, Du Bei, Wei Ruihua
Аннотация
BACKGROUND: The surging prevalence of paediatric myopia has overwhelmed clinical education services. While general-purpose large language models (LLMs) are increasingly used for medical information, they often lack the domain-specific precision required for specialised care. This study aimed to develop and evaluate Myopia in Kids Question-Answering (MiKid-QA), a paediatric myopia education question-answering model fine-tuned to support standardised patient education.
METHODS: We conducted a cross-sectional evaluation study. MiKid-QA was developed by fine-tuning the Qwen2.5-32B model using low-rank adaptation (LoRA) on curated professional datasets (2015-2025), including textbooks and expert consensus. Performance was assessed against DeepSeek and GPT-4 through a multicentre, single-blind expert evaluation of 25 standardised clinical scenarios. A panel of specialists rated responses across five dimensions (correctness, completeness, readability, helpfulness and safety) using a 5-point Likert scale.
RESULTS: MiKid-QA demonstrated superior automated performance compared with its base model (Bilingual Evaluation Understudy: 0.1423 vs 0.0967). In expert evaluations, MiKid-QA achieved significantly higher scores for correctness (4.43±0.63) and safety (4.38±0.59) compared with DeepSeek and GPT-4 (all p<0.001). While completeness and helpfulness were comparable (p>0.05), MiKid-QA produced more concise responses and lower reading difficulty than GPT-4 (p<0.001), suggesting a potentially favourable balance between information accuracy and patient accessibility in standardised educational scenarios.
CONCLUSION: Specialised fine-tuning on curated myopia-related datasets was associated with higher expert-rated correctness and safety in standardised myopia education scenarios. MiKid-QA may serve as an auxiliary educational tool to support patient communication in myopia care, although further external validation and prospective clinical studies are required before clinical deployment.
METHODS: We conducted a cross-sectional evaluation study. MiKid-QA was developed by fine-tuning the Qwen2.5-32B model using low-rank adaptation (LoRA) on curated professional datasets (2015-2025), including textbooks and expert consensus. Performance was assessed against DeepSeek and GPT-4 through a multicentre, single-blind expert evaluation of 25 standardised clinical scenarios. A panel of specialists rated responses across five dimensions (correctness, completeness, readability, helpfulness and safety) using a 5-point Likert scale.
RESULTS: MiKid-QA demonstrated superior automated performance compared with its base model (Bilingual Evaluation Understudy: 0.1423 vs 0.0967). In expert evaluations, MiKid-QA achieved significantly higher scores for correctness (4.43±0.63) and safety (4.38±0.59) compared with DeepSeek and GPT-4 (all p<0.001). While completeness and helpfulness were comparable (p>0.05), MiKid-QA produced more concise responses and lower reading difficulty than GPT-4 (p<0.001), suggesting a potentially favourable balance between information accuracy and patient accessibility in standardised educational scenarios.
CONCLUSION: Specialised fine-tuning on curated myopia-related datasets was associated with higher expert-rated correctness and safety in standardised myopia education scenarios. MiKid-QA may serve as an auxiliary educational tool to support patient communication in myopia care, although further external validation and prospective clinical studies are required before clinical deployment.