I-Corps: Detection of Strep Throat on a Telehealth Visit using Artificial Intelligence (AI) and Smartphone Images

这个I-Corps项目的更广泛的影响/商业潜力是远程保健技术的发展，将护理提供模式从面对面转变为虚拟。尽管最近有大量的医学人工智能（AI）文献，但很少有AI形式从理论背景转化为临床实践。该技术通过新颖的深度学习技术增强了AI性能，评估了AI医学测试在临床实践中的可用性和可解释性，并测量了AI测试的使用是否会改变抗生素处方实践。这项研究可能会扩展临床实践中使用的基于AI的临床决策支持软件的知识。改善获得虚拟医疗保健的机会可以影响那些最脆弱的人，例如服务不足的人，缺乏交通工具的人，残疾人或患有慢性病的人。例如，如果4000万次链球菌性咽喉炎的就诊都是在远程医疗中进行的，而不是在急诊室或紧急护理中进行的，那么健康保险公司每年可以节省40 - 120亿美元。这个I-Corps项目是基于开发一种新颖的深度学习算法，该算法使用智能手机的咽喉视频来预测链球菌性咽喉炎。目前，通过使用扁桃体的模式识别和基本的机器学习，从智能手机照片中检测链球菌咽喉是可行的。然而，目前的技术使用高质量的单帧图像，在移动的患者中获取这些图像是不切实际的，或者它们使用患者在家中不容易访问的硬件。一种新的方法是使用智能手机视频进行机器学习。此外，还有链球菌咽喉特有的视觉变化，如扁桃体发红或化脓，医生只能以55%的准确率识别。该技术应用这些视觉变化的模式识别，以区分链球菌性咽炎（链球菌性咽喉炎）和病毒性咽炎的智能手机图像。该奖项反映了NSF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 telehealth technology that shifts the care delivery model from in-person to virtual. Despite the wealth of recent medical artificial intelligence (AI) literature, few forms of AI translate from a theoretical context into clinical practice. The proposed technology enhances AI performance with novel, deep learning techniques, evaluates the usability and interpretability of an AI medical test in clinical practice, and measures whether use of the AI test alters antibiotic prescribing practices. This research may expand the knowledge around AI-based clinical decision support software used in clinical practice. Improved access to virtual healthcare can impact those who are most vulnerable, such as the underserved, those who lack transportation, those who have disabilities, or tjpse who have chronic illness. For example, if each of the 40 million strep throat visits were performed on telehealth instead of emergency rooms or urgent cares, health insurers could save $4-12 billion/year.This I-Corps project is based on the development of a novel, deep learning algorithm that predicts strep throat using a smartphone video of the throat. It is currently feasible to detect strep throat from a smartphone photo by using pattern recognition of the tonsils and basic machine learning. However, the current technologies use high quality, single frame images that are impractical to acquire in a moving patient, or they use hardware not easily accessible to patients at home. A new approach is to use smartphone video for machine learning. Furthermore, there are visual changes unique to strep throat, such as redness or pus on the tonsils, that doctors can only identify with 55% accuracy. This technology applies pattern recognition of these visual changes to differentiate smartphone images of streptococcal pharyngitis (strep throat) from viral pharyngitis.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.