應用於體積型醫療影像之無失真混合式壓縮編碼

Using Hybrid Coders for Lossless Compression on Volumetric Medical Images

隨著科技的進步及發展，醫療影像數位化技術在醫療領域有相當程度的革新與發展。由於數位醫學造影設備越來越普遍，數位醫學影像的數量也隨之急劇增加，使得所需的儲存空間呈爆炸性的成長。壓縮的原理是去除影像中的冗餘性來減少資料容量，以有效降低影像所佔之儲存空間，因此，若能根據被壓縮資料本身的特性來修改壓縮流程，將使壓縮效果更為顯著。體積型醫學影像中，相鄰的二維影像切面(slice)間通常具有高相似的像素分佈、邊緣分佈或特徵，適當利用這些特性，有效地去除相鄰影格間的冗餘性，能使壓縮效率能再進一步的提升。SPIHT (Set Partitioning in Hierarchical Trees)編碼法為一種先利用小波轉換集中能量，再利用轉換後的係數特性進行壓縮的方法。它擁有不錯的壓縮率，且複雜度不高，為目前熱門之壓縮法之一。本研究提出一改良之SPIHT精鍊過程，使其壓縮效果更加提升，並分別測試2D SPIHT、3D SPIHT、motion+SPIHT（動態補償後再使用2D SPIHT）等三種以SPIHT為核心之無失真壓縮流程，測試三者應用於體積型醫療影像上之效果。結果顯示，在切面間距較小時，motion+SPIHT可得到最好之壓縮率，而在間距大時，2D SPIHT可以有較好的效果，至於3D SPIHT之效果比起其他兩者則略為遜色，但不管是哪個方法，都能有高於JPEG-LS之表現，可以減少儲存電腦斷層影像所需的空間。

In the past ten years, medical industry has made conspicuous improvement in the technology of digital medical imaging. Because the widespread use of medical imaging devices, the size of medical images archive increases rapidly. The principle of compression is to remove the redundancy in images to reduce the image data size. More efficient compression can be achieved by adapting compression scheme according to the characteristics of the images. Neighboring slices in volumetric medical images usually have similar contents. If the redundancy between two slices is removed, the compression will be attained accordingly. SPIHT (Set Partitioning in Hierarchical Trees) is a compression algorithm based on discrete wavelet transform technique. Compression is achieved by utilizing the characteristics in coefficients after the transform. It is one of the popular schemes because of its low complexity and nice compression ratio. In this study, an improved refining procedure in SPIHT is proposed. Three schemes: 2D SPIHT, 3D SPIHT, and motion+SPIHT (using SPIHT after Motion Compensation), were applied on volumetric medical images to find the compression ratios. Results show, when applied on images with small slice thickness, motion+SPIHT can achieve better compression than the other two schemes. However, 2D SPIHT outperforms the other two on images with large slice thickness. 3D SPIHT shows no advantage to the others in either condition. All of the three schemes attain better compression ratios than the JPEG-LS does. The proposed schemes can reduce the required space for storing CT (computed tomography) images.