Information Processing for Medical Image Understanding

这个研究项目可以用一个简单的问题来描述：“计算机科学如何提高医生对疾病过程的理解？”“在拟议的工作中，复杂信息的新颖显示将为探索代谢过程和指导外科医生切除所有癌性肿瘤提供工具;在骨科，改进的数学建模将用于解决医学图像中骨关节炎的高度可变表现。 这项提案是利用申请人在计算方面的专业知识和相当丰富的手术经验，给出新的理论结果和实施，将直接改善加拿大人的健康。新的显示器将包括质谱数据的听觉呈现。在皇后医院，我们有几种仪器可以不同地捕获组织的质谱图像，或者可以在手术电刀切割组织时捕获数据流。这些数据是高度多光谱的，很难自动分类。相反，我建议将大量数据转换为可听输出。这将使研究组织样本的病理学家或在手术室中做出实时决策的外科医生能够真正听到癌症的声音。能够听到样本中每个细胞的代谢产物将是癌症研究组织分析的世界领先突破。数学建模将包括在高维数学流形上执行计算的方法。首先，我们将基于流形开发关节炎关节的统计模型;接下来，我们将使用统计模型来改进3D图像（如CT或MRI扫描）中关节炎关节的自动识别。关节炎是一系列疾病，影响到几乎每个加拿大家庭，因此，改善的方式来描述它将有助于分期治疗这种衰弱的条件。24名高素质的人员将接受多学科培训，为他们在生物医学计算的职业生涯做好准备。我们的毕业生受到医学院，研究生院和信息技术产业部门的高度追捧。

This research program can be phrased as a simple question: ``How can computer science improve a physician's understanding of disease processes?'' In the proposed work, novel displays of complex information will provide tools for exploring metabolic processes and guiding a surgeon to excise all of a cancerous tumor; in orthopedics, improved mathematical modeling will be used to address the highly variable presentation of osteoarthritis in medical images. This proposal is to use the applicant's expertise in computing, and considerable experience in surgery, to give new theoretical results and implementations that will directly improve the health of Canadians.The novel displays will include the auditory presentation of data from mass spectrometry. At Queen's, we have several instruments that can variously capture mass-spectrometry images of tissue, or can capture a data stream as a surgical electric knife slices through tissue. These data are highly multi-spectral and are very difficult to classify automatically. Instead, I propose to transform the mass data to an audible output. This will let a pathologist studying a tissue sample, or a surgeon making real-time decisions in an operating room, to literally hear the sound of cancer. Being able to hear the metabolic products of each cell in a sample would be a world-leading breakthrough in the analysis of tissues for cancer research.The mathematical modeling will include ways to perform computations on high-dimension mathematical manifolds. First, we will develop statistical models of arthritic joints based on the manifolds; next, we will use the statistical models to improve the automatic recognition of arthritic joints in 3D images such as CT or MRI scans. Arthritis is a constellation of diseases that affects nearly every Canadian family, so improved ways of describing it will help the staging of treatments for this debilitating condition.Twenty-four highly qualified personnel will receive multidisciplinary training to prepare them for careers in biomedical computing. Our graduates are highly sought after by medical schools, graduate schools, and information-technology industry sectors.