膠囊內視鏡用之腸胃道影像處理整合晶片設計

Abstract

本計畫目標在於設計可供膠囊內視鏡用之影像處理晶片。在膠囊內視鏡的應用中，延長電池的用電 壽命、降低資料的傳輸量、維持影像品質為我們的主要考量。然而這三個考量卻往往無法兼顧。在 有限的電源供應下，在長達三小時流程中持續耗電將嚴重限制影像處理能力與成像解析度。利用時 尚的影像壓縮技術固然可以因高壓縮比而大幅降低資料的傳輸量，但是大量的運算處理卻造成電池 電能的短缺。而利用犧牲影像品質以減少運算量的快速演算法卻可能造成醫療影像的失真。因此， 我們將從擷取影像時機管理、精簡壓縮演算法、設計功率意識組織三方面著手。首先，我們將利用 低功耗/高動態範圍仿視網膜晶片在微弱光源下偵測病變發生點，據以啟動耗電量較大的影像擷取與 壓縮電路。其次，再就人體腸胃道擷取之影像特色將現有影像壓縮技術加以簡化，同時結合特殊掃 描方式之影像感測器發展低運算量的壓縮演算法，並使此低運算量影像壓縮演算法具可倍乘/除性。 最後，根據有機電池耗電模式發展功率意識硬體組織以延長電池壽命。功率意識硬體組織可根據電 池電能的狀態適度犧牲影像壓縮品質換取電池的使用時間。

The objective of this project is to develop a power-aware image processor for capsule endoscopy or swallowable imaging capsules. The image processor consists three parts: disease target detector, image sensor, and image compressor. In applications of capsule endoscopy, it is imperative to consider battery life/performance trade-offs. Applying state-of-the-art video compression techniques may significantly reduce the image bit rate by their high compression ratio, but they all require intensive computation and consume much power from battery. There are many fast compression algorithms for reducing computation load; however, they may result in distoration of original image, which is not good for the use of medical care. Furthermore, the imaging capsules will travel in alimentary canal for three hours. Enabling all functions of the image processor over the three-hour trip will limit the functionality and resolution of image processing because of insufficient battery power. Hence, this project aims on three objectives: detection of diagnosing target, ultra-low-power gastrointestinal (GI) image compression, and power-aware architecture. First of all, this project will apply the retinal processing circuit for the disease target detection. The retinal processing circuit features high dynamic range, low power, and high sensitivity to motion object. These features make the capsule endoscopy able to operate in the dim canal initially, with part of lights. When sensing the targets, the endoscopy will turn on all the lights and the image processing circuits. Secondly, this project will simplify traditional video compression algorithms as per the characteristics of GI image. With a specific scanning on the CMOS sensor, we will propose a scalable compression algorithm for the later power-stepping technique. Finally, we will develop a power-aware architecture for battery-life extension. The power-aware architecture is an architecture that can properly reduce the computation load as the battery status changed while the qulity degradation is little. In the project, we will consider not only the minimization of average power dissipation but transient characteristics of power dissipation, such as peak power and power gradient or differential.