Department of Computer Science and Engineering, University of Bologna, Bologna, Italy.
Department of Biomedical and Neuromotor Sciences, University of Bologna, Bologna, Italy; IRCCS Istituto delle Scienze Neurologiche di Bologna, Bologna, Italy.
Sleep Med. 2018 Dec;52:7-13. doi: 10.1016/j.sleep.2018.07.018. Epub 2018 Aug 10.
Although being the most specific symptom of narcolepsy type 1 (NT1), cataplexy is currently investigated by clinical interview only, with potential diagnostic pitfalls. Our study aimed at testing the accuracy of an automatic video detection of cataplexy in NT1 patients vs. non-cataplectic subjects undergoing a standardized test with emotional stimulation.
Fifteen drug-naive NT1 patients and 15 age- and sex-balanced non-cataplectic subjects underwent a standardized video recording procedure including emotional stimulation causing laughter. Video recordings were visually inspected by human scorers to detect three typical cataplexy facial motor patterns (ptosis, mouth opening and head drop), and then analysed by SHIATSU (Semantic-based HIearchical Automatic Tagging of videos by Segmentation using cUts). Expert-based and automatic attack detection was compared in NT1 patients and non-cataplectic subjects.
All NT1 patients and none of the non-cataplectic subjects displayed cataplexy during emotional stimulation. Automatic detection correlated well with experts' assessments in NT1 with an overall accuracy of 81%. In non-cataplectic subjects, automatic detection falsely identified cataplexy in two out of 15 (13.3%) subjects who showed active eyes closure during intense laughter as a confounder with ptosis.
Automatic cataplexy detection by applying SHIATSU to a standardized test for video documentation of cataplexy is feasible, with an overall accuracy of 81% compared to human examiners. Further studies are warranted to enlarge the range of elementary motor patterns detected, analyse their temporal/spatial relations and quantify cataplexy for diagnostic purposes.
尽管猝倒症是 1 型发作性睡病（NT1）最具特异性的症状，但目前仅通过临床访谈进行调查，存在潜在的诊断陷阱。我们的研究旨在测试自动视频检测 NT1 患者猝倒症与非猝倒患者接受情绪刺激标准化测试的准确性。
15 名未经药物治疗的 NT1 患者和 15 名年龄和性别匹配的非猝倒患者接受了标准化视频记录程序，包括引起笑声的情绪刺激。视频记录由人工评分员进行视觉检查，以检测三种典型的猝倒面部运动模式（上睑下垂、口张开和头部下垂），然后由 SHIATSU（基于语义的 HIearchical 自动视频分段标记使用 cUts）进行分析。比较了基于专家和自动检测在 NT1 患者和非猝倒患者中的表现。
所有 NT1 患者在情绪刺激时均出现猝倒，而非猝倒患者无一例出现。自动检测与 NT1 患者专家评估相关性良好，总体准确率为 81%。在非猝倒患者中，自动检测错误地识别出两名（13.3%）在剧烈大笑时表现出主动闭眼的受试者发生猝倒，这是与上睑下垂混淆的因素。
通过将 SHIATSU 应用于标准化测试以自动检测猝倒来记录视频，是可行的，与人工检查相比，总体准确率为 81%。需要进一步研究以扩大检测到的基本运动模式的范围，分析其时间/空间关系，并量化猝倒以用于诊断目的。
