基于人工智能的重型新型冠状病毒感染恢复期患者肺部CT影像变化分析

AI-based analysis of lung CT imaging changes in severe COVID-19 patients during recovery

目的新型冠状病毒感染(coronavirus disease 19,COVID-19)对全球健康产生了重大影响,尤其是部分恢复期患者存在长新冠综合征(long-coronavirus disease,Long-COVID)症状及肺部影像学异常,然而肺部CT影像的长期动态变化特征尚未完全明确。本研究通过人工智能(artificial intelligence,AI)定量技术,动态分析重症COVID-19患者在1年内的肺部CT影像变化特征,为Long-COVID的管理和治疗提供参考。方法纳入干细胞治疗COVID-19队列中对照组(接受安慰剂及标准治疗方案)的58例受试者,收集基线及第1、3、6、9、12个月共6个时间点的临床及影像资料,应用AI定量技术测量病灶的质量、体积及密度值,分析病变的位置、范围及成分比例等,描绘重症COVID-19患者1年内的肺部影像学演变特征。结果绝大部分患者双肺受累,少数为单侧受累。不同肺叶受累程度不一,右肺下叶感染最严重,感染区域体积及体积比最大,随时间恢复正常的患者最少。基于CT密度值亨氏单位划分病灶成分,磨玻璃影较为常见,实变体积占全肺体积的比例较低。随时间推移,感染区域整体表现为密度下降、体积及体积比缩小的趋势,尤其在0~6个月期间逐渐减小,6个月后轻微回升。至第12个月时,所有患者的肺部CT影像仍残留异常表现。结论本研究利用AI定量技术发现,重症COVID-19患者肺部病变在恢复期逐渐吸收,但1年后仍未完全恢复正常。建议开展更长时间的随访研究,以监测COVID-19患者的肺部病变转归及特征,这对研究患者远期预后及Long-COVID相关机制具有重要意义。

Objective The Coronavirus Disease 2019(COVID-19)has significantly impacted global health,particularly as some recovering patients exhibit symptoms of Long-COVID and pulmonary imaging abnormalities.However,the long-term dynamic characteristics of lung CT imaging in these patients remain insufficiently understood.This study employs artificial intelligence(AI)quantitative techniques to dynamically analyze lung CT changes over the course of one year in patients with severe COVID-19,providing insights for the management and treatment of Long-COVID.Methods A total of 58 subjects from the placebo arm of a stem cell therapy COVID-19 cohort were included.Clinical and imaging data were collected at 6 time points:baseline,and at 1,3,6,9,and 12 months.AI-driven quantitative techniques were used to assess lesion mass,volume,and density,and to analyze the location,extent,and component ratios of lesions,depicting the radiographic evolution of lung changes in severe COVID-19 patients over one year.Results Most patients exhibited bilateral lung involvement,with a few showing unilateral involvement.The extent of lung involvement varied across lobes,with the right lower lobe being the most severely affected,having the largest infection volume and volume ratio,and the fewest patients recovering over time.Based on CT density values in Hounsfield units,ground-glass opacities were more frequently observed,while consolidation volumes constituted a smaller proportion of the total lung volume.Over time,the infected regions demonstrated a general trend of decreasing density,volume,and volume ratio,with a gradual decline from 0 to 6 months,followed by a slight increase after 6 months.At 12 months,residual lung abnormalities were still present in all patients’CT scans.Conclusions This study,utilizing AI-based quantitative techniques,revealed that lung lesions in severe COVID-19 patients gradually resolved during the recovery period.However,abnormal lung imaging findings persisted in all patients after one year.Prolonged follow-up studies are necessary to monitor the progression and characteristics of lung lesions in COVID-19 patients,as this is crucial for understanding longterm prognosis and mechanisms related to Long-COVID.