Machine Learning for Fibre-Bundle Endomicroscopy Systems

虽然我们对医学和生理学的理解最终必须牢牢扎根于自然科学，但临床医生的大部分知识及其诊断效果本质上是统计和经验的。虽然这种隐含或“直觉”的知识通常很难形式化，甚至很难表达，但应用统计学习（特别是所谓的“深度学习”）的最新进展使我们能够利用人类专家的专业知识，并最终自动化一些常规执行的较低级别的诊断工作。除了节省时间（这可以说是大多数现代临床环境中最稀缺的资源）之外，这些技术还可以为FLIM项目提出的诊断问题提供最先进的工程解决方案，如果没有人工智能的帮助，人类专家几乎无法获得这些解决方案。

While our understanding of Medicine and Physiology is and must ultimately be firmly rooted in the natural sciences, much of the knowledge of Clinicians and their resulting diagnostic efficacy is statistical and experiential in nature. While this kind of implicit or "intuitive" knowledge is often hard to formalize or even articulate, recent advances in applied statistical learning (particularly what is know as "deep learning") allow us to leverage the expertise of human experts and ultimately automate some of the lower-level diagnostic work the routinely perform. In addition to saving time, which is arguably the scarcest resource in most modern clinical environments, these techniques could power novel state-of-the-art engineering solutions to diagnostic problems such as that proposed by the FLIM project, which would be virtually inaccessible to human experts without the aid of artificial intelligence.