Computational methods for medical image analysis

这笔赠款的目的是为了扩大计算机视觉和机器学习方法在医学图像分析中的使用，用于两个具体的应用：(1)从超声心动图成像推断心肌纤维，(2)基于图像的癌症肿瘤外观的表征和量化。对于第一种应用，心肌纤维构成了心肌的底物，因此它们在心脏病学中具有重要意义。超声心动图是临床上使用最广泛的成像手段，能够从超声心动图推断心肌纤维的几何形状，这将允许创建针对患者的心脏模型，以促进非侵入性诊断和早期治疗。关于第二种应用，随着一些新的靶向药物疗法的出现，癌症患者的治疗选择在过去几年里急剧增加。然而，由于癌症的性质极其多样化，只有少数患者可能从任何一种特定的靶向治疗中完全受益。因此，重要的是能够在治疗过程的早期预测哪些患者最有可能从特定的治疗中受益。这可以通过非侵入性成像和图像分析算法来实现，图像分析算法可以根据肿瘤的外观特征预测治疗反应。

The goal of the present grant is to extend the use of computer vision and machine learning methods in medical image analysis for two specific applications: (1) the inference of cardiac muscle fibers from echocardiography imaging, and (2) the image-based characterization and quantification of cancerous tumor appearance. Regarding the first application, the cardiac muscle fibers form the substrate of the cardiac muscle, and as such they are of major importance in cardiology. The ability to infer the geometry of cardiac muscle fibers from echocardiography, the most widely used imaging modality in the clinic, will allow the creation of patient-specific cardiac models to facilitate non-invasive diagnosis and early treatment. With regards to the second application, there has been a dramatic increase in therapeutic options for patients with cancer over the last few years, with the advent of a number of novel targeted drug therapies. However, given the extremely diverse nature of cancer, only a small number of patients are likely to fully benefit from any one particular targeted treatment. It is therefore important to be able to predict, early in the course of treatment, which patients are most likely to benefit from a particular treatment. This can be achieved with non-invasive imaging, together with algorithms for image analysis that can predict response to treatment based on features of tumor appearance.