Modified Distance Regularized Level Set Segmentation Based Analysis for Kidney Stone Detection

Abstract

The abnormalities of the kidney can be identified by ultrasound imaging. The kidney may have structural abnormalities like kidney swelling, change in its position and appearance. Kidney abnormality may also arise due to the formation of stones, cysts, cancerous cells, congenital anomalies, blockage of urine etc. For surgical operations it is very important to identify the exact and accurate location of stone in the kidney. The ultrasound images are of low contrast and contain speckle noise. This makes the detection of kidney abnormalities rather challenging task. Thus preprocessing of ultrasound images is carried out to remove speckle noise. In preprocessing, first image restoration is done to reduce speckle noise then it is applied to Gabor filter for smoothening. Next the resultant image is enhanced using histogram equalization. The preprocessed ultrasound image is segmented using distance regularized level set segmentation (DR-LSS), since it yields better results. It uses a two-step splitting methods to iteratively solve the DR-LSS equation, first step is iterating LSS equation, and then solving the Sign distance equation. The second step is to regularize the level set function which is the obtained from first step for better stability. The DR is included for LSS for eliminating of anti-leakages on image boundary. The DR-LSS does not require any expensive re-initialization and it is very high speed of operation. The RD-LSS results are compared with distance regularized level set evolution DRLSE1, DRLSE2 and DRLSE3. Extracted region of the kidney after segmentation is applied to Symlets (Sym12), Biorthogonal (bio3.7, bio3.9 & bio4.4) and Daubechies (Db12) lifting scheme wavelet subbands to extract energy levels. These energy level gives an indication about presence of stone in that particular location which significantly vary from that of normal energy level. These energy levels are trained by Multilayer Perceptron (MLP) and Back Propagation (BP) ANN to identify the type of stone with an accuracy of 98.6%.