医学
经颅多普勒
卷积神经网络
血管痉挛
蛛网膜下腔出血
大脑中动脉
人工智能
人工神经网络
接收机工作特性
模式识别(心理学)
心脏病学
缺血
内科学
计算机科学
作者
Yong‐Jae Kim,DongWoon Go,Ilwoong Kim,Joo‐Hwan Kim,Yosub Park
出处
期刊:Stroke
[Ovid Technologies (Wolters Kluwer)]
日期:2023-02-01
卷期号:54 (Suppl_1)
标识
DOI:10.1161/str.54.suppl_1.tmp107
摘要
Purpose: Detection of cerebral vasospasm (CV) following aneurysmal subarachnoid hemorrhage (SAH) is imperative as this is one of the main determinants of delayed cerebral ischemia and poor neurological outcome. A transcranial Doppler (TCD) is a noninvasive device that can effectively detect CV. However, accurate interpretation of a TCD-based CV is time-consuming and demands expertise, and thus is underused in clinical practice. Artificial Intelligence (AI) models can assist in interpretation, reducing subjectivity, and speeding up the detection of CV. Methods: We conducted an automated method using neural networks to detect CV from TCD audio signal, obtained as short sequences of 3-5 seconds. A convolutional layer-based neural network is constructed that receives preprocessed two-dimensional data and estimates the peak systolic, end-diastolic, and mean blood flow velocity. Data were classified into three categories: normal, CV, and hyperemia. Results: 2,292 audio files, from 173 test data of 19 patients in the SAH registry were analyzed. Of these, the training dataset was comprised of 1,727 doppler wave files, the remaining 565 files were evaluated for validation using the proposed classifier. The ground truth was vascular neurologist’s previously reported TCD diagnosis. The accuracy for CV was 0.91, the precision was 0.90 and the sensitivity was 0.95. The area under the curve of CV discrimination was 0.97 and the f1 score was 0.93. Conclusion: The results show that the deep learning method can be used for accurate analysis and interpretation of CV based on audio signal analysis of TCD spectra. This novel AI-assisted interpretation may facilitate the diagnostic accuracy of TCD for the detection of CV.
科研通智能强力驱动
Strongly Powered by AbleSci AI