FULLY AUTOMATED CYST SEGMENTATION IN ULTRASOUND IMAGES OF KIDNEY

Cyst is one of the most common lesions in kidney and ultrasound imaging is appropriate tool for detecting these lesions. This study develops an automated approach for cyst segmentation in kidney’s ultrasound images. The approach includes three steps: initially, ultrasound image is transformed under a special function derived from Gibbs joint probability function. This transform suppresses noise and discriminates cyst and other tissues. Next, transformed image is decomposed to its low resolution component. Segmentation, morphological operations and coarse boundary detection (performed in low resolution) determines the initial contour employed in the final step. In last step, precise edge detection is performed in unity resolution using active contours model. Proposed approach is designed such that it overcomes noise, imaging artifacts and handles multi cyst cases. Coarse segmentation and then fine boundary extraction is an efficient scheme since segmentation is performed in low resolution (where SNR is relatively high) and lesion boundary is extracted precisely in high resolution (where details available).