Segmentation and Feature Extraction of Ultrasound Images by Modified Level Set Method and Chain-VESE Methods Using SRAD Filter

Abstract: Cancer is one of the most dreaded disease for mankind, if tumour are detected at initial stages then this disease can be completely cured, there are so many models in diagnosis of cancer one of these model is diagnosis through Ultrasound Images. So this paper proposes a model to segment the abnormality present in the Ultrasound image such as tumour or the lesion or the calculi. This model can be proposed for diagnosing and segmenting kidney calculi, lesions in ultrasound images. Compared to other methods of diagnosing tumours Ultrasound is more effective and easy to implement, less complicated and less in cost compared to CT scan and other methods. Segmentation of ultrasound images presents a unique challenge because these images contain strong speckle noise. There are many models for segmenting lesions from Ultrasound images but they are specified only to some organs like only to breast cancer or to prostrate cancer or only to the liver images but proposed model can be used to segment all Ultrasound B-mode images. So proposed model is tried on five different sets of Ultrasound images like kidney, liver and uterus to check the robustness of the model. The proposed model first the ultrasound image is developed from an organ then it is applied to SRAD filter to reduce the speckle and retain the edges of the images along with lesion part then we perform Automatic Thresholding is performed by Otsu algorithm followed by initializing Active contours by level set method and at last we segment the lesion by using Chan Vese model. Results are compared with various theoretical threshold values with the experimental values are closely approximated by the mean of zeroth and first order cumulative methods for texture features like area, solidity, perimeter. This method is simple compared to other methods and can be used for any B – mode Ultrasound images.

Keywords: Ultrasound Images, Fully Automatic Segmentation, Core Area, Adaptive Threshold, Prostrate Segmentation, Breast Lesion