Automated Ultrasound Image Segmentation Using AI: A Step Toward Non-invasive Kidney Disease Monitoring

Abstract

Non-invasive kidney evaluation through ultrasound imaging is quite common, but there is always a challenge in manual analysis of the kidney boundaries because of noise, brightness variation, and dependence on operators. The paper will examine a basic threshold-based segmentation algorithm to improve kidney boundaries in ultrasound images and their performance under different levels of image quality. Twenty-five kidney ultrasound images were pre-processed under the standardization and noise-reduction steps, and an automated boundary was further created through pixel-intensity thresholding. The standard of comparison was made of manual boundaries. The findings showed that the automated process had a close correspondence with manual contours in the majority of the cases, especially with the images that had moderate to high clarity. Comparison of the clarity groups revealed that there was a slight deviation of good quality images, moderate deviation in quality scanners, and higher variation amongst areas where noise or shadowing was evident. Morphological analysis also confirmed that the automated result did not cause significant changes in the general kidney anatomy, besides the fact that certain areas that had weak contrast had little differences. Also, the process of segmentation improved the visualization of structures by decreasing the effect of speckle interference and accentuating structure boundaries. On the whole, it can be concluded that an easy-to-use threshold-based segmentation method can offer a good and understandable kidney boundary extraction that can serve as a practical alternative to routine monitoring and clinical decision support, particularly in resource-limited environments.