A novel approach to ultrasound image thresholding using phase gradients

Abstract

Apart from being simple and cost-effective compared to other imaging techniques, ultrasound (US) imaging provides a non-invasive, ionization-free approach for diagnosis of tumours. However, US images are inherent with blurring and granular speckle noise which poses as a problem for effective diagnosis. The images also contain intensity inhomogeneities which make detection of tumours (similar to surrounding tissues) difficult. The commonly used segmentation approach is thresholding with Otsu thresholding being mostly used. However, due to intensity of inhomogeneities, normal non-tumour regions are also categorized as tumour regions in standard Otsu method. In this paper, we adapt a new method for thresholding of ultrasound images using phase gradients. Phase gradients have been used in microscopy techniques to enhance the visibility of low contrast structures within the cell body. As a pre-processing step, the image is enhanced using neutrosophic image enhancement and despeckled using shearlets. Neutrosophic approach is used considering the fuzzy nature of ultrasound images. The enhanced despeckled image is thresholded using an adaptive Otsu thresholding where the constraints for thresholding is derived also from texture, gradient, phase and phase gradient apart from mean and between class variance. The performance was evaluated using metrics like SSIM, MSE, SNR and accuracy revealing that the proposed method shows better results than the conventional Otsu thresholding. The proposed method was tested on US images with different echogenecities (hypoechoic, hyperechoic, anechoic, isoechoic) and showed promising results in identifying the tumour regions.