Lee Do-Kyeong, Choi Jae-Sung, Choi Seong-Jun, Choi Min-Hyung, Hong Min
Department of Software Convergence, Soonchunhyang University, Asan 31538, Republic of Korea.
Department of Internal Medicine, Cheonan Hospital, College of Medicine, Soonchunhyang University, Cheonan 31151, Republic of Korea.
Diagnostics (Basel). 2025 Jan 29;15(3):313. doi: 10.3390/diagnostics15030313.
This study proposes a classification system for predicting chronic obstructive pulmonary disease (COPD) patients and non-patients based on image and text data. This study measured the respiratory volume based on thermal images, stored the respiratory data, and derived features related to respiratory patterns, including the total respiratory volume, average distance between expirations, average distance between inspirations, and total respiratory rate. The data for each feature were stored in text format. The four features saved as text were scaled using Z-score normalization and expressed as scores through weighted summation. These scores were compared to a threshold based on the ROC curve values, classifying participants as patients if the score exceeded the threshold and as non-patients if it fell below. The proposed method achieved an accuracy of 82.5%. To validate the proposed approach, precision, recall, and F1-score were utilized, confirming the high classification performance of the model. The results of this study demonstrate the potential for future applications in non-contact medical examinations and diagnoses of respiratory diseases.
本研究提出了一种基于图像和文本数据预测慢性阻塞性肺疾病（COPD）患者和非患者的分类系统。本研究基于热图像测量呼吸量，存储呼吸数据，并推导与呼吸模式相关的特征，包括总呼吸量、呼气之间的平均距离、吸气之间的平均距离以及总呼吸频率。每个特征的数据以文本格式存储。以文本形式保存的这四个特征使用Z分数归一化进行缩放，并通过加权求和表示为分数。将这些分数与基于ROC曲线值的阈值进行比较，如果分数超过阈值，则将参与者分类为患者，如果分数低于阈值，则分类为非患者。所提出的方法实现了82.5%的准确率。为了验证所提出的方法，使用了精确率、召回率和F1分数，证实了该模型的高分类性能。本研究结果证明了该方法在非接触式医学检查和呼吸系统疾病诊断中的潜在应用前景。
