基于人工智能的肠道黏膜观察质量评估系统研究

An artificial intelligence‑based evaluation system for the quality of intestinal mucosal observation

摘要:

目的 建立基于人工智能的全面衡量肠道黏膜观察质量的监控系统，并探索该系统评分与腺瘤检出率（adenoma detection rate，ADR）之间的关系。方法 整合黏膜暴露程度评估模型、肠道准备评估模型、退镜速度监测模型3个模型，形成肠道黏膜观察质量评估系统（MQnet）。MQnet评分（0~3分）由黏膜暴露程度评分（0~1分）、肠道准备评分（0~1分）、退镜速度评分（0~1分）相加得到。回顾性分析来自武汉大学人民医院2020年7月1日至10月15日的854例肠镜受试者的859个视频资料，计算每例次肠镜的MQnet评分。使用Spearman相关性分析，评估MQnet评分和ADR之间的关系。结果 MQnet统计的评分段为2.0~<2.1分、2.1~<2.2分、2.2~<2.3分、2.3~<2.4分、2.4~<2.5分、2.5~<2.6分6个分数段，每个分数段对应的肠镜例次数分别为50、109、150、223、191和88，每个分数段对应的ADR分别为18.0%（9/50）、21.1%（23/109）、20.7%（31/150）、22.4%（50/223）、27.7%（53/191）和28.4%（25/88）。MQnet评分与ADR之间存在显著的正相关（Spearman系数为0.943，P<0.010）。结论 MQnet评分通过三个维度反映了肠道黏膜的观察质量，其与ADR之间呈现正相关的趋势，可用于量化评估肠镜检查质量。

Abstract:

Objective To establish an artificial intelligence-based evaluation system to comprehensively assess the quality of intestinal mucosal observation, and to explore the relationship between the score of the system and adenoma detection rate (ADR). Methods The intestinal mucosal observation quality assessment system (MQnet) was constructed by integrating the mucosal exposure assessment model, bowel preparation assessment model, and colonoscopy withdrawal speed monitoring model. MQnet score (0-3 points) was composed by adding mucosal exposure score (0-1 points), bowel preparation score (0-1 points), and withdrawal speed score (0-1 points). Data of 859 videos of 854 colonoscopy subjects at Renmin Hospital of Wuhan University from July 1st to October 15th 2020 were retrospectively analyzed. MQnet score of each colonoscopy was calculated and Spearman correlation analysis was conducted to assess the relationship between the MQnet score and ADR. Results The calculated MQnet score segments were 6 score bands of 2.0-<2.1, 2.1-<2.2, 2.2-<2.3, 2.3-<2.4, 2.4-<2.5, and 2.5-<2.6, with the number of colonoscopies corresponding to each band being 50, 109, 150, 223, 191, and 88, and with ADR corresponding to each band being 18.0% (9/50), 21.1% (23/109), 20.7% (31/150), 22.4% (50/223), 27.7% (53/191), and 28.4% (25/88), respectively. There was a significant positive correlationship between MQnet score and ADR (Spearman''s coefficient of 0.943, P<0.010). Conclusion MQnet score reflects the quality of intestinal mucosal observation through 3 dimensions, showing a positive correlationship with ADR, which can be used to quantitatively evaluate the quality of colonoscopy.