在肝癌的預後及治療計劃中,肝癌病理切片的分級是非常重要的。然而在不同的病理醫師之間,對於切片的分級結果都相當的主觀,因為現今的分級方式大多都是以視覺直接做判斷,所以不同人員之間判斷的準則也都不盡相同。為了避免這種分級不一致的狀況,藉由電腦來進行客觀的量化分級是必需的。 病理醫師在進行切片影像分級時,主要的判斷依據就是影像中細胞核的各種紋理特徵,所以要讓電腦可以進行正確的量化分析,最重要的先決條件就是病理切片影像中所有的細胞核都要非常的清楚。但是顯微鏡在同一個景深拍攝時,無法將範圍中的所有細胞核都清楚的拍攝下來,或是在拍攝過程中對焦錯誤,也會造成影像中的細胞變得模糊,而這些模糊的細胞核影像在電腦進行細胞核切割時將有可能導致形狀切割錯誤而影響最後的分級結果。因此,「一張所有細胞都拍攝清楚的切片影像」在利用電腦進行客觀量化分級時是非常有用的。 在本碩士論文中,我們提出一個影像融合的方法,它是以離散小波分解的聚焦測量(focus measure)方法為基礎,從多張不同景深所拍攝的切片影像中,將相對應的細胞取出清楚的部份,重新融合成一張包含所有清楚細胞的影像,以供電腦進行正確的肝癌病理切片影像分級。在我們的實驗中,我們將會證明經過融合的影像會比原來單焦距影像清晰,而且能保留其原來應有的影像特徵性質,使得在自動化分割和分級時能得到更好的結果。
The grading of pathological biopsy is very important in prognosis and treatment planning for hepatocellular carcinoma. Nevertheless, the grading results interpreted by pathologists are very subject to interobserver and intraobserver variability. Therefore, providing a quantitative analysis by machine vision is thus necessary. However, the cells on the biopsy are not all in the some depth of focus under the microscope. Small variance of focus may cause some of cells in captured image become a blur. These cells may not be segmented from image or segmented with incorrect shape by the machine and thus affect the grading results. Consequently, an “all-in-focus image” is very useful to the HCC grading performed by the machine. In this paper, we proposed an image fusion approach based on the wavelet-based focus measure to fuse two images with different depth of focus into one image, which contains much more in-depth focus cells. In our experiments, we demonstrated that the fused images not only provide clear appearance of cells but also higher accuracy of grading than original images.