Research Article
Implementation of Deep Learning in 3D Liver Image Segmentation
@INPROCEEDINGS{10.1007/978-3-032-34048-1_3, author={Endah Kristiani and Ronald Kartika and Chao-Tung Yang}, title={Implementation of Deep Learning in 3D Liver Image Segmentation}, proceedings={Smart Grid and Internet of Things. 9th EAI International Conference, SGIoT 2025, Taichung, Taiwan, December 6--7, 2025, Proceedings}, proceedings_a={SGIOT}, year={2026}, month={10}, keywords={3D Liver Segmentation U-NET Deep Learning}, doi={10.1007/978-3-032-34048-1_3} }- Endah Kristiani
Ronald Kartika
Chao-Tung Yang
Year: 2026
Implementation of Deep Learning in 3D Liver Image Segmentation
SGIOT
Springer
DOI: 10.1007/978-3-032-34048-1_3
Abstract
Accurate segmentation of liver tumors in 3D medical imaging is essential for effective diagnosis and treatment planning. Conventional 2D imaging often fails to capture the full volumetric context required for precise tumor assessment, creating potential risks for diagnostic errors. To address this, we implemented a U-Net-based deep learning framework capable of processing volumetric data to improve anatomical understanding. Using the Medical Segmentation Decathlon dataset (NIfTI format), we applied a robust pipeline involving standardizing, padding, and augmenting 3D data. The model was trained using the Adam optimizer with Binary Cross-Entropy Loss. Our results demonstrate that the U-Net architecture successfully distinguishes pathological tissue, achieving rapid convergence with a validation accuracy of approximately 98% by epoch 100. Although the average Intersection over Union (IoU) was 73.53%, the model showed strong stability without overfitting, confirming its potential for assisting in automated 3D liver segmentation tasks.