A U-Mamba-DLKA Framework with Artificial Rabbits Optimization for Automated Liver CT Image Segmentation

Effective liver segmentation within abdominal computed tomography (CT) scans is an essential requirement for clinical tasks including automated diagnostics, hepatic condition evaluation, and therapeutic progress tracking.However, automatic liver segmentation becomes challenging due to the substantial variations in liver shape and size, indistinct organ boundaries, and the presence of adjacent abdominal structures.Although existing convolutional neural network (CNN)-based and hybrid segmentation models have demonstrated significant performance, many exhibit limited capability in simultaneously capturing long-range contextual dependencies, adaptively refining spatial features, and efficiently optimizing training hyperparameters.To overcome these limitations, this study proposes an optimized liver CT image segmentation framework that integrates a U-Mamba encoder, Deformable Large Kernel Attention (DLKA), and a U-Net decoder within the segmentation architecture, while Artificial Rabbits Optimization (ARO) is employed for automated hyperparameter optimization.The proposed framework employs a hierarchical feature learning strategy in which the U-Mamba encoder combines convolutional operations with selective state-space modelling (SSM) to extract both local anatomical characteristics and long-range contextual dependencies.Subsequently, the DLKA module adaptively refines spatial and contextual feature representations through deformable large-kernel attention to improve contextual understanding and boundary localization, while the U-Net decoder progressively reconstructs high-resolution liver segmentation masks.ARO automatically determines the optimal training hyperparameters, including the learning rate, dropout rate, weight decay, and batch size.The framework was evaluated on the annotated Liver Tumor Segmentation (LiTS) dataset and achieved a Dice Similarity Coefficient (DSC) of 96.50%, an Intersection over Union (IoU) of 93.24%, a Precision of 97.12%, a Recall of 95.89%, and a 95th percentile Hausdorff Distance (HD95) of 4.10 mm.Furthermore, the ablation study demonstrated the contribution of the DLKA module and of systematic hyperparameter optimization, the five-fold cross-validation analysis indicated the stability of the selected model configuration, and the comparative analysis with conventional, Mamba-based, and liver-specific segmentation models showed competitive performance under an optimization protocol in which only the proposed framework received hyperparameter search.

Authors

Institutions

Publication Details

Journal
International journal of intelligent engineering and systems
Published
2026-11-17
DOI
https://doi.org/10.22266/ijies2026.1231.62
Primary Topic
Medical Image Segmentation Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A U-Mamba-DLKA Framework with Artificial Rabbits Optimization for Automated Liver CT Image Segmentation

Janisha A, R Mathusoothana S Kumar
1 links
International journal of intelligent engineering and systems
Medical Image Segmentation Techniques
article

A U-Mamba-DLKA Framework with Artificial Rabbits Optimization for Automated Liver CT Image Segmentation

Janisha A, R Mathusoothana S Kumar
article en

Abstract

Effective liver segmentation within abdominal computed tomography (CT) scans is an essential requirement for clinical tasks including automated diagnostics, hepatic condition evaluation, and therapeutic progress tracking.However, automatic liver segmentation becomes challenging due to the substantial variations in liver shape and size, indistinct organ boundaries, and the presence of adjacent abdominal structures.Although existing convolutional neural network (CNN)-based and hybrid segmentation models have demonstrated significant performance, many exhibit limited capability in simultaneously capturing long-range contextual dependencies, adaptively refining spatial features, and efficiently optimizing training hyperparameters.To overcome these limitations, this study proposes an optimized liver CT image segmentation framework that integrates a U-Mamba encoder, Deformable Large Kernel Attention (DLKA), and a U-Net decoder within the segmentation architecture, while Artificial Rabbits Optimization (ARO) is employed for automated hyperparameter optimization.The proposed framework employs a hierarchical feature learning strategy in which the U-Mamba encoder combines convolutional operations with selective state-space modelling (SSM) to extract both local anatomical characteristics and long-range contextual dependencies.Subsequently, the DLKA module adaptively refines spatial and contextual feature representations through deformable large-kernel attention to improve contextual understanding and boundary localization, while the U-Net decoder progressively reconstructs high-resolution liver segmentation masks.ARO automatically determines the optimal training hyperparameters, including the learning rate, dropout rate, weight decay, and batch size.The framework was evaluated on the annotated Liver Tumor Segmentation (LiTS) dataset and achieved a Dice Similarity Coefficient (DSC) of 96.50%, an Intersection over Union (IoU) of 93.24%, a Precision of 97.12%, a Recall of 95.89%, and a 95th percentile Hausdorff Distance (HD95) of 4.10 mm.Furthermore, the ablation study demonstrated the contribution of the DLKA module and of systematic hyperparameter optimization, the five-fold cross-validation analysis indicated the stability of the selected model configuration, and the comparative analysis with conventional, Mamba-based, and liver-specific segmentation models showed competitive performance under an optimization protocol in which only the proposed framework received hyperparameter search.

International journal of intelligent engineering and systemsVol. 19(12)
Noorul Islam University (IN)
Medical Image Segmentation Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon