Modified Otsu thresholding based level set and local directional ternary pattern technique for liver tumor segmentation

In recent times, the liver tumors are one of the leading causes of death, hence automated segmentation of liver tumors helps physicians in early diagnosis and treatment options. In this paper, a novel segmentation technique is proposed for accurate segmentation of tumor regions from the liver Ultrasound images. Initially, liver Ultrasound images are collected from a real time dataset, which comprises of 105 liver metastases images. Then, label removal is accomplished by using binary thresholding and morphological operation to remove text from the liver Ultrasound images. Additionally, the quality of liver Ultrasound images is improved by applying contrast limited adaptive histogram equalization that improves original image contrast and preserves the image brightness. After image enhancement, Otsu thresholding based level set with enhanced edge indicator function and local directional ternary pattern technique is proposed for segmenting liver lesion/tumor region from the enhanced images. In the experimental phase, the proposed technique performance is validated in light of Matthews's correlation coefficient, Jaccard coefficient, Dice coefficient, accuracy, precision and f-score. The simulation result showed that the proposed technique achieved 99.43% of segmentation accuracy, which is 5.43% higher than the existing graph based approach.