Development of computer vision algorithms for tissue identification and segmentation in medical ultrasound images

超声是对患者更友好的医学成像方式之一。这是婴儿的首选方式，因为设备可以被带到婴儿床或保温箱，并对婴儿造成最小的干扰或副作用。同样，这种设备可以很容易地带到成年患者手中，而且重病患者也能很好地耐受这种手术。然而，超声图像含有斑点和噪声，这使得从这些图像中提取诊断信息比其他成像方式更具挑战性。该研究项目的目标是开发自动化软件工具，帮助医生从这些超声图像中提取相关的诊断信息。特别是开发一种用于识别和表征组织类型的自动化技术，以便最终能够可靠地区分健康组织和不健康组织。提高使用对患者更友好的成像方式的可行性，同时仍能获得相同或更好的临床结果，将对患者的健康和医疗保健成本产生重大影响。

Ultrasound is one of the more patient friendly medical imaging modalities. It is the modality of choice for infants as the equipment can be brought to the bassinette or incubator and administered with minimal disturbance or side effects to the infant. Similarly the equipment can easily be brought to adult patients and the procedure is well tolerated by the extremely ill. However, ultrasound images contain speckle and noise making the extraction of diagnostic information from these images more challenging than for other imaging modalities. The goal of this research project is to develop automated software tools that will help the physician extract relevant diagnostic information from these ultrasound images. In particular to develop an automated technique for the identification and characterization of tissue types so that ultimately healthy tissue can be reliably distinguished from unhealthy tissue. Improving the viability of using a more patient friendly imaging modality while still achieving the same or better clinical results will have a significant impact on patient health and health care costs.