An efficient segmentation technique for skeletal scintigraphy image based on sharpness index and salp swarm algorithm

Abstract

Automatic Segmentation of skeletal scintigraphy images is an essential step for diagnosing bone metastasis and allowing definitive treatment. However, automatic segmentation of skeletal scintigraphy images is a challenging task due to the variance of the intensity distribution in different bones of the skeleton. This paper presents a multi-threshold technique to segment the dark spots of scintigraphy images. It segments the skull, the trunk, and the lower limbs separately rather than the whole skeletal scintigraphy image. Firstly, a sharpness index for each of the three parts is evaluated, and then an optimal threshold is computed using the salp swarm algorithm (SSA) with a fitness function based on maximizing the tsallis entropy function for each part. The proposed technique is implemented and applied with several different measures to an Egyptian medical dataset collected from Menoufia University Hospital. It is a real dataset, not standard dataset, that is a significantly addition to the research. It has its problems and drawbacks, which we are working to improve such as the invisible dark parts resulting from different external factors. The proposed technique is compared to the state-of-the-art techniques after applying the collected skeletal scintigraphy dataset to these techniques. Experimental results show that the proposed technique achieves superior performance in almost metrics such as mean square error (MSE), peak signal to noise ratio (PSNR), precision, recall, accuracy, normalized absolute error (NAE), Jaccard index, dice coefficient, matthews correlation coefficient (MCC), F1-Score, and structural similarity index matrix (SSIM).