Brains behind the eyes: Interpreting Medical Images

动机：现代医学成像仪生成详细的图像，提供前所未有的内部解剖视图。然而，这些机器本质上是复杂的“相机”或“眼睛”。尽管它们能生成漂亮的图像，但它们无法量化或“理解”这些图像。临床医生通常在视觉和定性上检查这些图像，因为从这些图像中进行精确测量的工具尚未在临床中可用。开发“眼睛”背后的“大脑”，或者图像背后的智能算法，将原始成像数据转换为可用于检测疾病发作、自信地诊断疾病或定量监测疾病进展的测量值，是我研究计划的动力和长期目标。**重点：我的研究项目重点是开发一个计算解剖学框架，以挖掘与阿尔茨海默病（AD）等脑部疾病相关的生物标志物的3D脑磁共振（MR）图像。第二个重点是开发一种新的计算解剖学框架，用于解释视网膜的3D光学相干断层扫描（OCT）图像，以寻找与眼病（如年龄相关性黄斑变性（AMD）或青光眼）相关的生物标志物。由于视网膜是大脑的神经感觉延伸，来自视网膜图像的生物标志物不仅对改善视力保健有价值，而且在未来，也可能作为大脑生物标志物的替代品，并可能有助于诊断或监测脑部疾病的进展。**挑战：每个3D图像通常有数百万个样本和测量。感兴趣的信号，即由于疾病引起的变化，通常定位在构成图像的体素的子集中（例如记忆回路驻留的海马体，或视神经头，视神经纤维从眼睛流出）。对于依赖于第一个任务的所有下游测量来说，在大型3D体积图像中自动识别感兴趣区域（分割）是一项具有挑战性和重要的任务。一旦分割，从感兴趣区域（ROI）获取的每个形态测量向量都是高维空间中的一个点。将信号（由于疾病或状况引起的ROI变化）与“扩散”（或由于人群中ROI的正常变化而引起的可变性）区分开来是第二个关键挑战。**研究目标：提高自动分割、配准和测量的准确性，设计更好的可分离特征和分类器，从而发现新的定量生物标志物，用于疾病诊断和进展。**意义：脑部和眼部疾病是65岁以上加拿大人致残的主要原因。它们导致个人及其家庭的生活质量下降，每年给加拿大造成数十亿美元的损失。所提议的框架有可能填补定量生物医学图像解释技术的主要未满足需求，有助于更好地利用成像数据中的信息，节省宝贵的临床医生时间，从而以更低的成本提供更高质量的医疗保健。**进展：过去的Discovery资助周期对我的实验室来说是非常富有成效的，有25篇高影响力的期刊出版物和17篇同行评审的会议出版物。我被授予著名的Michael Smith基金会健康研究学者奖和BC省专业工程师和地球科学家协会的功勋成就奖，以表彰我在设计临床应用的计算工具方面的贡献。基于这些优势，我创建了一个令人兴奋的研究项目，致力于定量生物医学图像解释的新发现，在过去的周期中，我指导了36名学员。

Motivation: Modern medical imagers generate detailed images providing an unprecedented view of the internal anatomy. However, these machines are essentially sophisticated `cameras' or `eyes'. Though they generate pretty pictures, they cannot quantify or `understand' these images. The clinicians typically examine these images visually and qualitatively since tools for making accurate measurements from these images are not yet available in the clinic. Developing the `brains' behind the `eyes', or the intelligent algorithms behind the images that can convert raw imaging data into measurements that can be used to detect the onset of disease, diagnose a disease with confidence, or to quantitatively monitor disease progression is the motivation and the long term goal of my research program. **Focus: My research program focuses on developing a computational anatomy framework to mine the 3D brain magnetic resonance (MR) images for biomarkers related to brain diseases such as Alzheimer's disease (AD). A second focus is to develop a novel computational anatomy framework for interpreting 3D optical coherence tomography (OCT) images of the retina for biomarkers related to eye diseases such as age-related macular degeneration (AMD) or glaucoma. Since the retina is a neurosensory extension of the brain, biomarkers from retinal images are not only valuable for improving vision care, but in the future, may also serve as surrogates for brain biomarkers and potentially help in diagnosing or monitoring the progression of brain diseases.**Challenges: Each 3D image typically has millions of samples and measurements. The signal of interest i.e. the changes due to disease, are typically localized in a subset of the voxels that make an image (such as the hippocampus, where memory circuits reside, or the optic nerve head, where optic nerve fibers exit the eye). Automated identification of regions of interest (segmentation) in large 3D volumetric images is a challenging and important task for all downstream measurements depend on this first task. Once segmented, each vector of morphometric measurements taken from a region of interest (ROI) is a point in a high-dimensional space. Distinguishing the signal (the changes in a ROI due to a disease or condition) from the `spread', or variability due to normal variation in the ROI in the population, is a second key challenge. **Research goals: are to increase accuracy in automated segmentation, registration and measurements, design better separable features and classifiers leading to the discovery of novel quantitative biomarkers for disease diagnosis and progression.**Significance: Brain and eye diseases are the primary cause of disability among Canadians over the age of 65. They lead to poor quality of life for the individual and their family, and cost Canada billions of dollars each year. The proposed frameworks have the potential fill a major, unmet need in quantitative biomedical image interpretation technologies, help better utilize information present in imaging data, save valuable clinician time, and thus allow better quality health care to be delivered at lower cost. **Progress: The past Discovery grant cycle has been highly productive for my lab, with 25 high-impact journal publications and 17 peer-reviewed conference publications. I was awarded the prestigious Michael Smith Foundation for Health Research Scholar award and the Association of Professional Engineers & Geoscientists of BC's Meritorious Achievement award for my contributions to designing clinically applicable computational tools. Building on these strengths, I have created an exciting research program towards new discoveries in quantitative biomedical image interpretation in which I supervised 36 trainees in the past cycle.