Hounsfield Unit Based Multi-Threshold Approach for Liver Segmentation

The liver is an essential organ involved in many physiological processes. Diseases affecting its structure can lead to severe health issues and even death, making early diagnosis and analysis from medical images crucial. Accurate liver segmentation is important for applications such as radiotherapy, surgery simulation, and diagnosis. However, with the abundance of data available, manual segmentation by radiologists has become a time-consuming task. To address these challenges and improve segmentation accuracy, we introduce an automatic liver segmentation method for abdominal CT images. Unlike traditional methods that rely solely on a global threshold value, the proposed method applies multiple threshold values based on the Hounsfield unit (HU) scale to segment the liver more accurately. The novelty of this approach lies in its integration of HU-based tissue density quantification with multi-thresholding techniques for better liver region representation. The methodology involves several key steps: using the rescale slope and intercept to convert raw pixel values to Hounsfield units, determining optimal threshold values based on standard HU ranges, generating a binary mask, and applying post-processing techniques. The method demonstrated promising results, achieving a Dice similarity coefficient of 98.22%, a relative volume difference of 0.035%, and a structural similarity index of 98.40% on the CHAOS dataset. This approach advances medical image segmentation particularly in liver segmentation and holds significant potential for clinical applications.