教练理论：一套临床人工智能系统开发与验证理论体系

A theoretical system for the development and validation of the clinical artificial intelligence systems

摘要:

近年来，人工智能技术蓬勃发展，在图像识别、特征提取等方面的优势使之有机会被广泛应用于医学影像和医疗文书等领域，作为辅助诊疗活动的重要工具。然而，临床医学科学家往往并非人工智能领域的专家，在选择构建或评估模型时需要与计算机领域的工程师专家沟通合作，不充分的沟通或不恰当的合作模式会导致资源的利用效率降低，无法实现高效率跨学科合作。对于临床医师，人工智能模型像一个语言不通的学徒，临床医师无法将自身的理论与实践经验直接传授给学徒，而需要一位懂得模型语言的翻译官。本文旨在规范临床人工智能系统开发和验证流程，提出一种医工合作的新模式，以指导临床人工智能系统的开发与应用，构建具有较高诊疗质量、较广泛适用场景的人工智能辅助诊疗系统。本文提出了临床人工智能系统开发的8个阶段，其中，临床专家医师作为教练员（coach）的角色，计算机工程师作为翻译官（translator）的角色，人工智能模型作为学徒（apprentice）的角色，以教练-翻译-学徒的模式（coach⁃translator⁃apprentice）进行高质量临床人工智能系统的开发与验证。

Abstract:

In recent years, there has been a notable surge in the utilization of artificial intelligence (AI) technology, particularly in the domains of image recognition and feature extraction. This has led to the widespread adoption of AI in various fields, including medical imaging and text analysis, where it is now regarded as a valuable tool for facilitating diagnostic and therapeutic activities. However, clinical scientists often lack expertise in the domain of AI, necessitating collaboration with engineers and computer science experts in the development and evaluation of models. Insufficient communication or inappropriate modes of collaboration may result in the inefficient utilization of resources. From the perspective of the clinician, the AI model is like an apprentice that is unable to communicate effectively in the desired language. It is not possible for the clinician to transfer their theoretical and practical experience to the apprentice directly, thus, a translator who is familiar with the language of the model is required. The objective of this theory is to establish a standardized process for the development and validation of the clinical AI systems, and to propose a novel model of medical-industrial cooperation to guide the development and validation of clinical AI systems. This theory presents an eight-stage process. In this process, the clinician plays the role of a coach, the computer engineer acts as a translator, and the AI model is an apprentice. The development and validation of high-quality clinical AI systems can be conducted in accordance with a COACH-TRANSLATOR-APPRENTICE paradigm.