Jeong Sangwoon, Pyo Hongryull, Park Won, Han Youngyih
Department of Health Sciences and Technology, SAIHST, Sungkyunkwan University, Seoul 06355, Republic of Korea.
Department of Radiation Oncology, Samsung Medical Center, Seoul 06355, Republic of Korea.
Cancers (Basel). 2024 May 22;16(11):1964. doi: 10.3390/cancers16111964.
This study aimed to predict stress in patients using artificial intelligence (AI) from biological signals and verify the effect of stress on respiratory irregularity. We measured 123 cases in 41 patients and calculated stress scores with seven stress-related features derived from heart-rate variability. The distribution and trends of stress scores across the treatment period were analyzed. Before-treatment information was used to predict the stress features during treatment. AI models included both non-pretrained (decision tree, random forest, support vector machine, long short-term memory (LSTM), and transformer) and pretrained (ChatGPT) models. Performance was evaluated using 10-fold cross-validation, exact match ratio, accuracy, recall, precision, and F1 score. Respiratory irregularities were calculated in phase and amplitude and analyzed for correlation with stress score. Over 90% of the patients experienced stress during radiation therapy. LSTM and prompt engineering GPT4.0 had the highest accuracy (feature classification, LSTM: 0.703, GPT4.0: 0.659; stress classification, LSTM: 0.846, GPT4.0: 0.769). A 10% increase in stress score was associated with a 0.286 higher phase irregularity ( < 0.025). Our research pioneers the use of AI and biological signals for stress prediction in patients undergoing radiation therapy, potentially identifying those needing psychological support and suggesting methods to improve radiotherapy effectiveness through stress management.
本研究旨在利用人工智能（AI）从生物信号预测患者的应激状态，并验证应激对呼吸不规则的影响。我们对41例患者中的123个病例进行了测量，并利用从心率变异性得出的七个与应激相关的特征计算应激分数。分析了整个治疗期间应激分数的分布和趋势。利用治疗前的信息来预测治疗期间的应激特征。AI模型包括非预训练模型（决策树、随机森林、支持向量机、长短期记忆网络（LSTM）和Transformer）和预训练模型（ChatGPT）。使用10折交叉验证、精确匹配率、准确率、召回率、精确率和F1分数来评估性能。计算了呼吸不规则的相位和幅度，并分析其与应激分数的相关性。超过90%的患者在放射治疗期间经历了应激。LSTM和提示工程GPT4.0的准确率最高（特征分类，LSTM：0.703，GPT4.0：0.659；应激分类，LSTM：0.846，GPT4.0：0.769）。应激分数增加10%与相位不规则性增加0.286相关（<0.025）。我们的研究率先利用AI和生物信号对接受放射治疗的患者进行应激预测，有可能识别出那些需要心理支持的患者，并提出通过应激管理提高放射治疗效果的方法。
