Date of Publication :25th June 2024

Abstract:The process of defining an organ of interest for volumetric or morphological study is known as segmentation. This is usually done on MRI scans or CT (computed tomography scans). The liver's highly changeable shape and close closeness to other organs make it one of the hardest organs to section. Furthermore, the liver can develop a variety of diseases that alter its architecture, density, or signal intensity. Accurate segmentation of the liver in medical imaging is crucial for precise diagnostics and effective treatment planning.This research introduces an innovative approach for liver segmentation using DuckNet, an advanced convolutional neural network designed explicitly for tasks involving medical image segmentation like liver segmentation. The Liver Duke Dataset, a comprehensive repository of 2046 MRI scans of the abdominal region encompassing a diverse range of cirrhotic liver cases, serves as a robust benchmark for evaluating DuckNet's segmentation performance. Using hierarchical features, the model accurately delineates the liver from surrounding organs in the abdominal region, demonstrating high precision. Precise segmentation of the liver is vital int automating the task of segmenting the liver from abdominal MRI scans. The training regimen involves optimizing DuckNet using a combination of liver images and corresponding ground truth masks. Results from comprehensive experiments highlight the effectiveness of DuckNet in accurately segmenting the liver. We used many metrics like pixel-wise accuracy in segmentation, jaccard similarity coefficient. If we talk about segmentation tasks then it is often used to get an estimate of the overlap between the ground truth region and the region segmented by our model. Comparative analyses with existing segmentation methods underscore DuckNet's superior performance on the Duke Dataset, positioning it as a robust tool for clinical applications. This research contributes to the advancement of liver segmentation methodologies, showcasing DuckNet's effectiveness on the challenging Liver Duke Dataset. The proposed approach holds great promise for enhancing diagnostic accuracy, facilitating treatment planning, and elevating overall patient care within the realm of hepatic imaging.

Reference :