Journal of obstetrics and gynaecology : the journal of the Institute of Obstetrics and Gynaecology · 2026
Evaluating AI in postoperative rehabilitation of axillary web syndrome after breast cancer surgery: ChatGPT vs DeepSeek.
СтатьяСравнительное исследованиеОценочное исследованиеНазвание и аннотация: Перевод в очереди
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.
Cheng Qun, Lou Weiwei, Mou Gue, Liu Yali
Аннотация
OBJECTIVE: To systematically compare two generative artificial intelligence (AI) models, ChatGPT and DeepSeek in the rehabilitation management for axillary web syndrome (AWS) following breast cancer surgery; to assess their potential value in supporting clinical rehabilitation decision-making.
METHODS: Five virtual cases, based on real clinical scenarios and representing varying levels of complexity and functional impairment, were constructed. Three rehabilitation medicine specialists independently evaluated the models' responses using a standardised scoring rubric. Assessment dimensions included diagnostic accuracy, comprehensiveness of evaluation, individualisation of treatment planning, safety and risk awareness, evidence-based support, and clarity of logic. Additional analyses examined clinical reasoning depth, error patterns, and differences in response style.
RESULTS: Both models demonstrated high inter-rater reliability. For overall clinical competency scores, DeepSeek performed significantly better than ChatGPT. DeepSeek showed superior performance in diagnostic accuracy (P = 0.006), comprehensiveness of evaluation (P = 0.034), and individualisation of therapeutic planning (P = 0.005). DeepSeek maintained robust performance across cases of varying complexity, with particularly strong advantages in diagnostically challenging cases and those involving concomitant lymphedema. ChatGPT more frequently exhibited generalised treatment recommendations and omissions in differential diagnoses, whereas DeepSeek, despite occasionally provided recommendations with lower levels of evidence, had a lower overall error rate. DeepSeek generated more actionable and context-specific responses, often including concrete exercise prescriptions and technical details. It also outperformed ChatGPT in reasoning chain completeness, depth of evidence justification, and mechanistic explanation rationality.
CONCLUSION: In this exploratory pilot study based on a limited number of virtual cases, DeepSeek demonstrates relatively better performance than ChatGPT in AWS rehabilitation decision support. However, findings should be interpreted cautiously, and further validation in real-world clinical settings is required.
METHODS: Five virtual cases, based on real clinical scenarios and representing varying levels of complexity and functional impairment, were constructed. Three rehabilitation medicine specialists independently evaluated the models' responses using a standardised scoring rubric. Assessment dimensions included diagnostic accuracy, comprehensiveness of evaluation, individualisation of treatment planning, safety and risk awareness, evidence-based support, and clarity of logic. Additional analyses examined clinical reasoning depth, error patterns, and differences in response style.
RESULTS: Both models demonstrated high inter-rater reliability. For overall clinical competency scores, DeepSeek performed significantly better than ChatGPT. DeepSeek showed superior performance in diagnostic accuracy (P = 0.006), comprehensiveness of evaluation (P = 0.034), and individualisation of therapeutic planning (P = 0.005). DeepSeek maintained robust performance across cases of varying complexity, with particularly strong advantages in diagnostically challenging cases and those involving concomitant lymphedema. ChatGPT more frequently exhibited generalised treatment recommendations and omissions in differential diagnoses, whereas DeepSeek, despite occasionally provided recommendations with lower levels of evidence, had a lower overall error rate. DeepSeek generated more actionable and context-specific responses, often including concrete exercise prescriptions and technical details. It also outperformed ChatGPT in reasoning chain completeness, depth of evidence justification, and mechanistic explanation rationality.
CONCLUSION: In this exploratory pilot study based on a limited number of virtual cases, DeepSeek demonstrates relatively better performance than ChatGPT in AWS rehabilitation decision support. However, findings should be interpreted cautiously, and further validation in real-world clinical settings is required.