Manuscript received May 26, 2024; revised July 24, 2024; accepted August 21, 2024; published January 17, 2025.

Abstract—Medical image segmentation is a crucial component in diagnostic and therapeutic processes, enabling precise analysis of anatomical structures to improve clinical outcomes. The liver is a particularly significant organ due to its multifaceted functions and its role in various physiological processes. Accurate segmentation of the liver from medical images is essential for disease diagnosis, treatment planning, and surgical interventions. This study used liver tumor segmentation data of 65 (Computed Tomography) CT scan datasets. However, challenges in 3D liver segmentation include heterogeneous textures, varying shapes, and proximity to neighboring structures, necessitating advanced techniques to enhance accuracy and efficiency. This study utilizes the dataset from the Liver Tumor Segmentation Challenge (LiTS), which includes contrast-enhanced abdominal CT scans from various clinical sites worldwide. The research proposes an optimized Convolutional Neural Networks (CNN) architecture, DeepLabV3+ (Proposed), which integrates atrous convolution with Atrous Spatial Pyramid Pooling (ASPP) and low-level feature fusion. The results indicate that the DeepLabV3+ model achieves the best performance, with an Intersection over Union (IoU) of 0.68 and a Dice Similarity Coefficient (DSC) of 0.84. The implication is that this model can significantly enhance liver segmentation accuracy in clinical practice, thereby improving the quality of patient diagnosis and treatment.

Keywords—3D liver segmentation, deep learning, DeepLabV3+, Atrous Spatial Pyramid Pooling (ASPP), LiTS dataset

Cite: Hersatoto Listiyono, Zuly Budiarso, and Agus Perdana Windarto, "An Optimized CNN Architecture for Accurate 3D Liver Segmentation in Medical Images," Journal of Image and Graphics, Vol. 13, No. 1, pp. 15-23, 2025.

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC-BY-4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.