An AI assistance tool to guide novice practitioners in the competent performance of flexible video laryngoscopy

摘要 Perceptron Health建议开发一款人工智能(AI)软件和图像处理 培训高级实践提供商(APP)以执行合格的柔性光纤的辅助工具 对患者进行喉镜检查(FFL)，提高他们的技能吸收。美国65.7%的县缺乏 执业耳鼻喉科医生(ENT)，这导致了基于地理位置的护理差异 地点。农村地区的人受到的影响最大。新的人工智能工具有可能增加 临床医生能够执行从13,000个Ent到近500,000个应用程序的喉镜检查，填补了危重护理 差距。该工具包将指导应用程序用户完成喉镜检查程序，以确保所有解剖结构 并充分捕获患者的任务。然后，可以远程查看和解释录音 内科医生，让他们专注于诊断和治疗。这款基于人工智能的产品将包括一幅图像 捕获指导系统以及跟踪成功的程序检查表和质量检查系统 捕捉关键解剖结构的可诊断视图。 Perceptron Health计划通过以下第一阶段目标评估该工具包的技术可行性：1. 开发一个原型软件工具包，通过喉镜检查程序提供指导；2.测试 原型工具的能力，以提高在人体模型上执行喉镜检查的能力；以及3.评估AI的 在预先录制的视频中概括人体解剖学的能力。 Perceptron的工具将通过创建一个从业者辅助工具来扩大患者接触FFL的机会 识别解剖结构，相对于解剖结构定位相机，并为 用户通过用户界面(UI)。在这个项目中产生的原型将需要新的开发 通过开发最新的能够从喉镜视频中分类图像的算法 ART卷积神经网络将允许将人工智能算法集成到喉镜中。这个 与传统方法相比，提出的算法可以提供实质性的改进，并将具有 应用于许多其他医学内窥镜检查环境(胃肠道、肺部和其他) 来处理喉镜检查视频中的图像。一旦完全开发，这项创新将允许非耳鼻喉科 临床医生在支持耳语和言语能力的同时，扩大他们的执业范围 病理学家进行更远程的护理，接触到更多的患者。该技术的其他潜在用户 包括急诊室医生和麻醉师。重要的是，这项拟议的技术有望改善健康 通过扩大获得专科护理的社会经济机会和减少治疗时间。

Abstract Perceptron Health proposes to develop an artificial intelligence (AI) software and image processing assistance tool that trains advanced practice providers (APPs) to perform a competent flexible fiberoptic laryngoscopy (FFL) on patients and improve uptake of their skills. With 65.7% of U.S. counties lacking a practicing ear, nose, and throat physician (ENT), this has led to a disparity in care based on geographical location. Those in rural areas are most impacted. The novel AI tool has the potential to increase the pool of clinicians capable of performing laryngoscopy from 13,000 ENTs to almost 500,000 APPs, filling a critical care gap. The toolkit will guide APP users through the laryngoscopy procedure to ensure all anatomical structures and patient tasks are sufficiently captured. A recording can then be reviewed and interpreted remotely by an ENT physician, allowing them to focus on diagnosis and treatment. The AI-based product will include an image capture guidance system as well as a procedure checklist and quality check system that tracks successful capture of diagnosable views of key anatomical structures. Perceptron Health plans to assess technical feasibility of the toolkit through the following Phase I Objectives: 1. Develop a prototype software toolkit that provides guidance through the laryngoscopy procedure; 2. Test the prototype tool’s capability to improve the ability to perform laryngoscopy on manikins; and 3. Assess the AI’s ability to generalize to human anatomy in pre-recorded video. Perceptron’s tool will expand patient access to FFLs via the creation of a practitioner assistance tool able to identify anatomical structures, localize the camera relative to anatomical structures, and provide guidance to the user through a user interface (UI). The prototype to be generated in this project will require the novel development of algorithms capable of classifying images from laryngoscopy videos through the development of state-of-the- art convolutional neural networks that will allow for the integration of AI algorithms into laryngoscopes. The proposed algorithms can provide substantial improvements relative to conventional approaches and will have application in numerous other medical endoscopy contexts (gastrointestinal, pulmonary, and others) in addition to processing images from laryngoscopy videos. Once fully developed, this innovation will allow non-ENT clinicians to expand their scope of practice while supporting the ability of both ENTs and speech language pathologists to perform more remote care and reach more patients. Other potential users of the technology include ER physicians and anesthesiologists. Importantly, the proposed technology is expected to improve health by expanding socioeconomic access to specialty care and decreasing time to treatment.