Explainable AI for diagnosing and treating cardiovascular disease

博士项目的目标：心脏病是全世界的头号杀手。人工智能模型可以自动诊断疾病，但它们缺乏解释力。该项目旨在开发用于心脏病学诊断和治疗计划的人工智能工具，可以向心脏病专家解释其决策。项目描述/背景：将人工智能（AI），特别是深度学习用于心脏病学的诊断和治疗计划是一个活跃的研究领域[1,2]。然而，虽然深度学习技术产生了令人印象深刻的结果，但仍然存在一个重大问题。通常，产生最准确结果的技术缺乏一个重要特征，这对临床接受新技术很重要：解释力。简而言之，大多数深度学习模型能够做出预测，但无法用人类可解释的术语解释预测是如何达成的。如果没有这样的解释，在许多应用中，临床医生将不愿意根据这些“黑箱”模型的建议做出临床决定。在这个项目中，我们专注于将图像（可能还有其他临床数据）作为输入的深度学习模型。从基于图像的深度学习模型中生成解释是一个重大挑战。在文献中，大多数对这种“可解释的机器学习”或“可解释的人工智能”的尝试都集中在两种方法中的一种：(1)尝试可视化“黑盒子”的内部，例如通过使用“显著性地图”来显示输入图像中对决策很重要的区域；(2)训练一个更简单的模型，在某种程度上可能更可解释。这两种方法在许多医学应用中可能都是不够的。例如，在心脏病学中，这是我们在这个项目中的重点，一个可以被心脏病专家接受的“解释”可能需要关于病理过程和/或概念的信息，如组织特性和电/机械激活模式。一个关键的挑战将是找到将模型的自动决策与“更高级别”的人类可解释概念联系起来的方法。在本项目中，我们将探讨如何在心衰患者诊断、分层和治疗计划的应用中建立这些联系。最近从计算机视觉文献中出现的一个有前途的领域是调查人类可解释概念对深度学习模型的重要性的方法。我们最近开始将这些方法应用于心脏病学，并取得了非常有希望的初步结果。其他有趣的探索途径包括将解释纳入训练目标[5]的方法，以及将人类（即临床医生）“置于”深度学习模型训练[6]的方法。这种类型的方法可以用来鼓励深度学习模型学习具有临床意义的特征，有效地在临床医生和深度学习模型之间建立对话。在这个领域有许多有趣的方向可以探索，而这些方向在医学领域是相对未触及的，并且具有很高的新颖性潜力。我们的最终目标是生产一个计算机辅助决策支持工具，以协助心脏病专家对心力衰竭患者进行分层并计划其治疗。这个工具就像一个“值得信赖的同事”或“第二读者”，心脏病专家可以与之接触，找到他们对困难病例的意见，以及这种意见背后的原因。这是一个雄心勃勃的目标，这个项目代表了这个旅程的第一部分，但如果成功，影响将是巨大的。

Aim of the PhD Project:Heart disease is number one killer worldwide.AI models can automatically diagnose disease, but they lack explanatory power.Project aims to develop AI tool for diagnosis and treatment planning in cardiology that can explain its decisions to cardiologists.Project Description / Background:The use of artificial intelligence (AI), and specifically deep learning, for diagnosis and treatment planning in cardiology is an active research area [1,2]. However, whilst deep learning techniques have produced impressive results, a significant problem remains. Often, the techniques that produce the most accurate results lack one important feature that is important for the clinical acceptance of new technology: explanatory power. Put simply, most deep learning models are able to make predictions but are not able to explain in human-interpretable terms how the prediction was arrived at. Without such explanations, in many applications clinicians will be reluctant to base clinical decisions upon recommendations from such "black-box" models. In this project we focus on deep learning models that take images (and possibly other clinical data) as input. Producing explanations from image-based deep learning models is a significant challenge. In the literature, most attempts at such "interpretable machine learning" or "explainable AI" have focused on one of two approaches: (1) try to visualise the inside of the "black box", e.g. by using "saliency maps" which show areas of the input image that were important in making the decision, (2) train a simpler model which may be more interpretable to some degree. Both of these approaches are likely to be inadequate in many medical applications. For example, in cardiology, which is our focus in this project, an "explanation" that will be acceptable to a cardiologist is likely to require information about pathological processes and/or concepts such as tissue properties and electrical/mechanical activation patterns. A key challenge will be to find ways of linking the model's automated decision with "higher level" human-interpretable concepts. In this project, we will investigate ways of making these links in the application of patient diagnosis, stratification and treatment planning in heart failure. One promising area that has recently emerged from the computer vision literature is the investigation of ways of querying the importance of human-interpretable concepts to deep learning models [3]. We have recently started to apply these methods in cardiology with highly promising initial results [4]. Other interesting avenues for exploration include methods that incorporate explanations into the training objective [5], as well as ways of putting humans (i.e. clinicians) "in the loop" of the training of deep learning models [6]. This type of approach could be used to encourage the deep learning model to learn features that are clinically meaningful, effectively creating a dialogue between clinicians and deep learning models. There are many intriguing directions to explore in this field which are relatively untouched in the medical domain, and the potential for novelty is high. Our ultimate aim is to produce a computer aided decision-support tool to assist cardiologists in stratifying patients with heart failure and planning its treatment. The tool would act like a "trusted colleague" or "second reader" that the cardiologist could engage with to find their opinion about difficult cases as well as the reasoning behind this opinion. This is a highly ambitious aim and this project represents the first part of this journey, but if successful the impact could be great.