Jha Manika, Gupta Richa, Saxena Rajiv
Department of Electronics and Communication Engineering, Jaypee Institute of Information Technology, 201309 Noida, India.
SN Comput Sci. 2023;4(1):89. doi: 10.1007/s42979-022-01526-x. Epub 2022 Dec 13.
The association of pulmonary fibrosis with COVID-19 patients has now been adequately acknowledged and caused a significant number of mortalities around the world. As automatic disease detection has now become a crucial assistant to clinicians to obtain fast and precise results, this study proposes an architecture based on an ensemble machine learning approach to detect COVID-19-associated pulmonary fibrosis. The paper discusses Extreme Gradient Boosting (XGBoost) and its tuned hyper-parameters to optimize the performance for the prediction of severe COVID-19 patients who developed pulmonary fibrosis after 90 days of hospital discharge. A dataset comprising Electronic Health Record (EHR) and corresponding High-resolution computed tomography (HRCT) images of chest of 1175 COVID-19 patients has been considered, which involves 725 pulmonary fibrosis cases and 450 normal lung cases. The experimental results achieved an accuracy of 98%, precision of 99% and sensitivity of 99%. The proposed model is the first in literature to help clinicians in keeping a record of severe COVID-19 cases for analyzing the risk of pulmonary fibrosis through EHRs and HRCT scans, leading to less chance of life-threatening conditions.
新冠肺炎患者与肺纤维化之间的关联现已得到充分认识，并在全球范围内导致了大量死亡。由于自动疾病检测现已成为临床医生获得快速准确结果的关键助手，本研究提出了一种基于集成机器学习方法的架构来检测与新冠肺炎相关的肺纤维化。本文讨论了极端梯度提升（XGBoost）及其调整后的超参数，以优化对出院90天后发生肺纤维化的重症新冠肺炎患者的预测性能。研究考虑了一个包含1175例新冠肺炎患者的电子健康记录（EHR）和相应胸部高分辨率计算机断层扫描（HRCT）图像的数据集，其中包括725例肺纤维化病例和450例正常肺部病例。实验结果的准确率达到98%，精确率为99%，灵敏度为99%。所提出的模型是文献中首个帮助临床医生通过EHR和HRCT扫描记录重症新冠肺炎病例以分析肺纤维化风险的模型，从而降低危及生命情况的发生几率。
