醫療儀器推動小組：臨床超音波診斷儀之開發與臨床評估─子計畫二：臨床應用軟體與超音波影像處理演算法之研發（III）

Abstract

臨床超音波診斷儀的開發，是否能受到廣 大的使用者歡迎，決定於兩類主要的因素： 亦即客觀及主觀的因素。所謂客觀的因素指 的是系統本身的規格與品質是否能與同級產 品相競爭。而主觀的因素則是系統所提供的 功能是否合乎一般醫師之需求。本子計劃的 主要目標即是要發展一套臨床應用軟體不僅 能符合臨床之真正需求，更提供俱智慧且有 善的操作功能以節省醫師的操作時間。 為了要使醫師能以最少的步驟達成其所需 的軟體功能，半自動或全自動的影像處理技 術是此應用軟體成功與否的關鍵所在。是 以，於此計畫中，除了撰寫應用軟體及其中 必要的人性化人機介面之外，重點將在於研 發適用於超音波影像的影像處理演算法。由 於超音波的影像品質比其他的醫學影像的品 質差了許多，傳統的影像處理方法並不完全 適用。是以在此計畫中，我們將發展新的影 像處理技術，希望能讓使用者輕鬆的完成所 需的分析工作。於第三年的計畫中，我們提 出一個新的形變模型，稱為離散型梯度流形 變模型。此模型以對比梯度與叢聚元件強化 微弱邊緣。且以離散型梯度流導引遠處的初 始輪廓至目標物的邊緣。此模型已成功的應 用於超音波胎兒影像的分割。

In the development of an ultrasound imaging system, whether it will become a popular system strongly depends on two factors, namely, objective and subjective factors. While the objective factor means whether the specification and the quality can be compete with other products of the same class, the subjective factor means whether the operation functions match most doctors' needs. The goal of this component project is to develop an application software with operation functions not only matching clinical needs but also being intelligent and user-friendly to minimize a doctor's operation time. In order to allow a doctor to complete an operation function with the minimal effort, semi-automatic and automatic image processing algorithms would be the key technology. Therefore, in this project, in addition to write the application software and its user-friendly man-machine interface, the emphasis would be on developing the image processing algorithms suitable for ultrasound images. Since the quality of an ultrasound image is much worse than that of other types of medical images, the conventional image processing techniques may not be very effective for an ultrasound image. Hencs, in this project, we have put a great effort on developing new image processing algorithms to deal with images for different applications, hoping that the user can easily complete their desired operation functions. In the third-year project, we have proposed a new deformable model, called Discrete Gradient Flow Deformable Model. The salient features of this model are to enhance the weak edges by using the contrast gradient and the connected components and guide the distant initial contour with the new image force, i.e., the Discrete Gradient Flow. This model has been verified by using a set of ultrasound fetus images.