SRMA-Mamba: Spatial Reverse Mamba Attention Network for Pathological Liver Segmentation in MRI Volumes

Jun Zeng, Yannan Huang, Elif Keles, Halil Ertugrul Aktas, Gorkem Durak, Nikhil Kumar Tomar, Quoc-Huy Trinh, Deepak Ranjan Nayak, Ulas Bagci, Debesh Jha · 2025

Read SRMA-Mamba: Spatial Reverse Mamba Attention Network for Pathological Liver Segmentation in MRI Volumes by Jun Zeng, Yannan Huang, Elif Keles, Halil Ertugrul Aktas, Gorkem Durak, Nikhil Kumar Tomar, Quoc-Huy Trinh, Deepak Ranjan Nayak, Ulas Bagci, Debesh Jha (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-12410.txt.

Introduction

Researchers present a deep learning network for outlining diseased livers in MRI scans in this 2025 paper. Cirrhosis is a late, dangerous stage of chronic liver disease. Catching it early and acting in time can cut deaths, yet the liver's complex shape and varied damage make its scans hard to read.

Past methods underuse the spatial layout inside a full 3D scan, which hurts both accuracy and trust. The new network, SRMA-Mamba, is built on the Mamba architecture, a fast sequence model. A key module scans selectively within liver tissue while weaving in anatomical knowledge. A spatial reverse attention step then sharpens the borders of the outlined regions. The aim is segmentation that follows the organ's real structure, so doctors can measure damage with more confidence.

Catalog record

CREATORJun Zeng, Yannan Huang, Elif Keles, Halil Ertugrul Aktas, Gorkem Durak, Nikhil Kumar Tomar, Quoc-Huy Trinh, Deepak Ranjan Nayak, Ulas Bagci, Debesh Jha
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-12410

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

SRMA-Mamba: Spatial Reverse Mamba Attention Network for Pathological Liver Segmentation in MRI Volumes Jun Zeng, Yannan Huang, Elif Keles, Halil Ertugrul Aktas, Gorkem Durak, Nikhil Kumar Tomar, Quoc-Huy Trinh, Deepak Ranjan Nayak, Ulas Bagci, Debesh Jha Abstract Liver Cirrhosis plays a critical role in the prognosis of chronic liver disease. Early detection and timely intervention are critical in significantly reducing mortality rates. However, the intricate anatomical architecture and diverse pathological changes of liver tissue complicate the accurate detection and characterization of lesions in clinical settings. Existing methods underutilize the spatial anatomical details in volumetric MRI data, thereby hindering their clinical effectiveness and explainability. To address this challenge, we introduce a novel Mamba-based network, SRMA-Mamba, designed to model the spatial relationships within the complex anatomical structures of MRI volumes. By integrating the Spatial Anatomy-Based Mamba module (SABMamba), SRMA-Mamba performs selective Mamba scans within liver cirrhotic tissues and combines anatomical information from the sagittal, coronal, and axial planes to construct a global spatial context representation, enabling efficient volumetric segmentation of pathological liver structures. Furthermore, we introduce the Spatial Reverse Attention module (SRMA), designed to progressively refine cirrhotic details in the segmentation map, utilizing both the coarse segmentation map and hierarchical encoding features. Extensive experiments demonstrate that SRMA-Mamba surpasses state-of-the-art methods, delivering exceptional performance in 3D pathological liver segmentation. Our code is available for public: https://github.com/JunZengz/SRMA-Mamba . Introduction Cirrhosis constitutes a significant global health burden, ranking as the 11th leading cause of mortality worldwide in 2019 ( Huang et al. 2023 ) . As the end-stage manifestation of progressive liver fibrosis, it remains a principal cause of death among individuals with chronic liver disease. Cirrhosis most commonly arises from chronic viral hepatitis, metabolic dysfunction-associated steatotic liver disease (MAFLD), and alcohol-related liver disease ( Huang et al. 2023 ) . A hallmark pathological feature is the disruption of normal hepatic architecture, replaced by regenerative nodules. The condition is frequently associated with sinusoidal portal hypertension, which constitutes a major risk factor for hepatocellular carcinoma ( Bartolomeo, Trerotoli, and Serio 2011 ) . Severe complications, including hepatic encephalopathy, ascites, and hepatorenal syndrome, frequently manifest during the late or decompensated phase of cirrhosis, indicating a substantial decline in patient survival ( Tapper and Parikh 2023 ) . Liver biopsy remains a pivotal procedure in the detection and diagnosis of liver diseases, particularly in assessing liver fibrosis and cirrhosis ( Chowdhury and Mehta 2023 ) . However, it is associated with inherent risks, including hemorrhage and infection, owing to its invasive nature. Magnetic resonance imaging (MRI), as a non-invasive diagnostic modality, offers superior soft tissue contrast, positioning it as an essential diagnostic modality for the identification of cirrhotic lesions and the assessment of liver pathology. Morphological alterations and variations in signal intensity within MRI volumes play a critical role in the diagnosis and staging of cirrhosis ( Tonan, Fujimoto, and Qayyum 2010 ) . CirrMRI600+ is a large-scale dataset of cirrhotic MRI volumes to develop automatic segmentation methods with deep learning ( Jha et al. 2025 ) . Computer-assisted segmentation enables rapid and precise localization of cirrhotic lesions, thereby enhancing clinical decision-making and optimizing treatment planning. Volumetric segmentation is a critical task in medical image analysis, aiming to delineate the region of interest (ROI) within volumetric data. Convolutional neural networks (CNNs) ( Çiçek et al. 2016 ; Milletari, Navab, and Ahmadi 2016 ; Oktay et al. 2018 ) for volumetric segmentation commonly utilize an encoder-decoder architecture. Due to the localized nature of convolutional kernels, CNNs struggle to capture long-range dependencies in the volumetric data. In contrast, transformers ( Cao et al. 2022 ; Hatamizadeh et al. 2022 ; Shaker et al. 2024 ) utilize self-attention mechanisms to model global relationships, enabling them to achieve significant success in volumetric segmentation. However, transformers incur substantial computational costs due to the quadratic complexity of the self-attention mechanism. Recently, state-space models (SSMs) ( Gu, Goel, and Ré 2021 ; Smith, Warrington, and Linderman 2022 ; Gu and Dao 2023 ) have demonstrated exceptional performance in capturing long-range dependencies. Mamba ( Gu and Dao 2023 ) employs a selection mechanism and a hardware-aware algorithm to model global dependencies, achieving state-of-the-art performance in linear-time sequence modeling with linear complexity. To extend Mamba into the vision domain, VMamba ( Liu et al. 2024 ) performs four distinct traversal path scans to achieve 2D global receptive field. Based on this foundation, Mamba-based methods ( Ruan, Li, and Xiang 2024 ; Wang et al. 2024 ; Zeng et al. 2025 ; Xing et al. 2024 ) have emerged in the field of medical image segmentation. RMAMamba ( Zeng et al. 2025 ) effectively segments pathological liver regions within MRI slices by employing a novel reverse mamba attention mechanism, yet fails to capture the intricate spatial dependencies inherent in volumetric MRI data. Segmamba ( Xing et al. 2024 ) takes the whole volumetric data as a linear sequence and explores the forward, reverse, and inter-slice features interaction. However, SegMamba exhibits limitations in capturing complex spatial anatomical relationships, as it linearizes volumetric data into a one-dimensional sequence, thereby disregarding the inherent spatial context, an essential factor for accurate interpretation in clinical applications. To address these issues, we propose SRMA-Mamba, to model the spatial relationships within the complex anatomical structures of MRI volumes for efficient volumetric pathological liver segmentation. The main contributions of the work are as follows: • We present a novel spatial reverse mamba attention network, SRMA-Mamba, for efficient volumetric pathological liver segmentation in MRI volumes. • We propose a novel Spatial Anatomy-Based Mamba (SABMamba) module to perform scans within multiple anatomical perspectives, including sagittal, coronal, and axial planes. Additionally, we introduce the Spatial Reverse Mamba Attention (SRMA) module, which progressively refines the segmentation map using hierarchical features and coarse segmentation map. • We introduce the Anatomy-Based Selective Scan (ABSS) module to model 3D visual feature representations, enabling direct processing of 3D volumetric data with low computational complexity. This module addresses the gap left by Vision Mamba in the 3D vision domain. • We conduct extensive experiments on the CirrMRI600+ T1W and T2W datasets and demonstrate that SRMAMamba achieves state-of-the-art performance in 3D pathological liver segmentation, surpassing existing volumetric segmentation methods. Figure 1: Overview of the proposed SRMA-Mamba architecture. The network utilizes an encoder-decoder architecture specifically designed for accurate volumetric segmentation of pathological liver regions in MRI volumes. In the encoder, the input MRI scan is initially divided into embeddings, reducing the resolution to half of the original size. In the following four encoding stages, the input feature at each stage is processed by an SABMamba block, generating multi-scale hierarchical feature maps. In the decoder, the feature maps from each stage, together with the coarse segmentation map from the previous stage, are input into the SRMA block. SRMA block progressively refines the segmentation results by combining feature optimization with the complement of boundary information. The network integrates the SABMamba encoder and SRMA blocks to achieve coarse-to-fine segmentation boundary refinement, resulting in enhanced segmentation accuracy and performance. Related work Volumetric Segmentation Volumetric (3D) segmentation is a fundamental challenge in medical imaging, enabling precise localization, delineation, and quantification of organs and tumors. This technique is indispensable for various clinical applications, including surgical planning, treatment monitoring, and the assessment of disease

Read the complete text