I-Corps: Artificial Intelligence-Based Clinical Decision Support for Acute Stroke Victims

I-Corps项目的更广泛影响/商业潜力是开发一种计算技术，以减少服务不足的医院急诊科误诊中风的发生率。这些医院可能没有专门的中风中心，或者接触中风专家的机会有限。急诊科医生面临的挑战是识别中风的不同症状，并将中风与多种症状重叠的疾病区分开来。漏诊或迟发中风会导致永久性残疾或本可避免的死亡，并给医院和患者带来不必要的治疗费用。除了提高中风诊断的准确性外，计算支持系统还可以提高诊断的速度，从而增加仅在中风症状发作后3小时内有效的时间敏感药物的使用。因此，快速准确地诊断中风可以改善中风患者的长期预后。这个I-Corps项目探索发展的翻译，以改善临床决策支持系统，以帮助急性中风诊断。该技术是一种实时系统，可以通过视频片段量化中风症状，如虚弱、面部下垂和不协调。建议的技术可以支持急诊科医生的诊断，而不需要有经验的中风专家在场。拟议的技术采用了一套经过验证和发布的特定机器学习算法，这些算法是在不断增长的中风患者接受神经系统检查的视频和录音数据集上训练的。研究结果表明，只需10秒钟的中风患者被动静坐的视频片段，该技术就能以80%的准确率检测偏瘫，超过了神经学家进行基于视频的中风评估的表现。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this I-Corps project is the development of a computational technology to reduce the occurrence of misdiagnosed strokes in the emergency departments of underserved hospitals. These hospitals may not have a specialized stroke center or may have limited access to stroke specialists. Emergency department physicians are challenged to identify the different symptoms of stroke and to distinguish stroke from a diverse range of diseases that have overlapping symptoms. Missed or late diagnosis of strokes result in permanent disabilities or otherwise preventable deaths and incur unnecessary treatment costs for both hospitals and patients. In addition to improving the accuracy of stroke diagnosis, a computational support system may increase the speed of diagnosis, thereby increasing the administration of time-sensitive drugs that are only effective within 3 hours of the onset of stoke symptoms. Diagnosing stroke quickly and accurately therefore improves long-term stroke patient outcomes.This I-Corps project explores the translation of developments to improve clinical decision support systems to aid acute stroke diagnosis. The technology is a real-time system that is able to quantify stroke symptoms like weakness, facial droop, and incoordination simply from video footage. The proposed technology may support diagnosis by emergency department physicians without requiring an experienced stroke specialist to be on hand. The proposed technology employs a particular set of validated and published machine learning algorithms trained on a growing dataset of video and audio recordings of stroke patients undergoing neurological exams. Results show that from just 10 seconds of video footage of stroke patients sitting passively at rest, the technology is able to detect hemiparesis with 80% accuracy, exceeding the performance of neurologists conducting a video-based assessment of stroke.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.