SBIR Phase I: Development of a Machine Learning System to Identify Streptococcal Pharyngitis with a Smartphone Image

这个小企业创新研究（SBIR）第一阶段项目的更广泛影响/商业潜力解决了远程保健缺乏即时远程医疗测试的问题。本项目可开发基于机器学习的链球菌性喉炎准确预测模型。该业务模式将基于人工智能（AI）的临床决策支持系统作为软件即服务订阅提供给紧急护理远程医疗服务。紧急护理和初级保健中所有远程保健护理点测试（链球菌性咽喉炎除外）的潜在市场总额为104亿美元。这一解决方案影响了抗生素的过度处方和卫生服务的经济性。目前，34%的咽炎儿童和75%的咽炎成人接受了不必要的抗生素治疗，而远程医疗的情况更糟10-21%。链球菌性咽喉炎的远程护理点预测可能会减少每年2200万美元的不必要抗生素费用，并减少耐药细菌的驱动因素。与急诊室或紧急护理机构相比，通过远程保健治疗咽炎可为患者每次临床就诊节省多达1-3小时，每次就诊可为健康保险公司节省多达100-400美元。这个小企业创新研究（SBIR）第一阶段项目通过结合智能手机图像分析和深度学习来推进机器学习领域。这些策略应用于数字健康中的一个新用例，即临床决策支持的远程筛查。技术上的挑战是开发一种预测模型，以达到临床采用可接受的敏感性和特异性，目标为80%（类似于快速抗原链球菌试验）。应对这一挑战的策略是增加数据集的大小，并尝试多个预测模型，直到达到目标性能。该项目还将包括设计一个认证系统，验证未经训练的患者记录的足够图像，并创建一个直观的用户界面，使患者能够一致记录。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 Small Business Innovation Research (SBIR) Phase I project addresses the lack of instant, remote medical tests for telehealth. This project could develop an accurate machine learning-based predictive model for strep throat. The business model delivers an artificial intelligence (AI)-based clinical decision support system as a Software as a Service subscription to urgent care telehealth services. The total addressable market for all telehealth point of care tests (beyond strep throat) in urgent care and primary care is $10.4 billion. This solution impacts antibiotic overprescribing and economics of health services. Currently, 34% of children and 75% of adults with pharyngitis receive unnecessary antibiotics, and this is 10-21% worse with telehealth. A remote point of care prediction for strep throat can potentially reduce the $22 million/year costs in unnecessary antibiotics and reduce drivers for drug-resistant bacteria. When pharyngitis is treated on telehealth it saves patients up to 1-3 hours per clinical visit and saves health insurance companies up to $100-400 per visit, compared to an emergency room or urgent care facility.This Small Business Innovation Research (SBIR) Phase I project advances the field of machine learning by combining smartphone image analysis and deep learning. These strategies are applied to a novel use case in digital health as remote screening for clinical decision support. The technical challenge is the development of a predictive model to achieve sensitivity and specificity acceptable for clinical adoption, at a target of 80% (similar to the rapid antigen strep test). The strategy to meet this challenge is to increase the size of the dataset and experiment with multiple prediction models until goal performance is achieved. The project will also include designing an authentication system that validates sufficient images as recorded by an untrained patient and creating an intuitive user interface that enables consistent recordings by patients.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.