다중과제 상호작용 적대학습 기반 간 종양 영상 통합 분석 네트워크

📝 Abstract

Liver tumor segmentation, dynamic enhancement regression, and classification are critical for clinical assessment and diagnosis. However, no prior work has attempted to achieve these tasks simultaneously in an end-to-end framework, primarily due to the lack of an effective framework that captures inter-task relevance for mutual improvement and the absence of a mechanism to extract dynamic MRI information effectively. To address these challenges, we propose the Multi-Task Interaction adversarial learning Network (MTI-Net), a novel integrated framework designed to tackle these tasks simultaneously. MTI-Net incorporates Multi-domain Information Entropy Fusion (MdIEF), which utilizes entropy-aware, high-frequency spectral information to effectively integrate features from both frequency and spectral domains, enhancing the extraction and utilization of dynamic MRI data. The network also introduces a task interaction module that establishes higher-order consistency between segmentation and regression, thus fostering inter-task synergy and improving overall performance. Additionally, we designed a novel taskdriven discriminator (TDD) to capture internal high-order relationships between tasks. For dynamic MRI information extraction, we employ a shallow Transformer network to perform positional encoding, which captures the relationships within dynamic MRI sequences. In experiments on a dataset of 238 subjects, MTI-Net demonstrates high performance across multiple tasks, indicating its strong potential for assisting in the clinical assessment of liver tumors. The code is available at: https://github.com/xiaojiao929/MTI-Net .

💡 Analysis

Liver tumor segmentation, dynamic enhancement regression, and classification are critical for clinical assessment and diagnosis. However, no prior work has attempted to achieve these tasks simultaneously in an end-to-end framework, primarily due to the lack of an effective framework that captures inter-task relevance for mutual improvement and the absence of a mechanism to extract dynamic MRI information effectively. To address these challenges, we propose the Multi-Task Interaction adversarial learning Network (MTI-Net), a novel integrated framework designed to tackle these tasks simultaneously. MTI-Net incorporates Multi-domain Information Entropy Fusion (MdIEF), which utilizes entropy-aware, high-frequency spectral information to effectively integrate features from both frequency and spectral domains, enhancing the extraction and utilization of dynamic MRI data. The network also introduces a task interaction module that establishes higher-order consistency between segmentation and regression, thus fostering inter-task synergy and improving overall performance. Additionally, we designed a novel taskdriven discriminator (TDD) to capture internal high-order relationships between tasks. For dynamic MRI information extraction, we employ a shallow Transformer network to perform positional encoding, which captures the relationships within dynamic MRI sequences. In experiments on a dataset of 238 subjects, MTI-Net demonstrates high performance across multiple tasks, indicating its strong potential for assisting in the clinical assessment of liver tumors. The code is available at: https://github.com/xiaojiao929/MTI-Net .

📄 Content

Adversarial Multi-Task Learning for Liver Tumor Segmentation, Dynamic Enhancement Regression, and Classification Xiaojiao Xiao1, Qinmin Vivian Hu1, Tae Hyun Kim2, and Guanghui Wang1 1 Department of Computer Science, Toronto Metropolitan University, Toronto, ON, Canada 2 Department of Computer Science, Hanyang University, Seoul, South Korea Abstract Liver tumor segmentation, dynamic enhancement regression, and classification are critical for clinical assessment and di- agnosis. However, no prior work has attempted to achieve these tasks simultaneously in an end-to-end framework, pri- marily due to the lack of an effective framework that captures inter-task relevance for mutual improvement and the absence of a mechanism to extract dynamic MRI information effec- tively. To address these challenges, we propose the Multi- Task Interaction adversarial learning Network (MTI-Net), a novel integrated framework designed to tackle these tasks si- multaneously. MTI-Net incorporates Multi-domain Informa- tion Entropy Fusion (MdIEF), which utilizes entropy-aware, high-frequency spectral information to effectively integrate features from both frequency and spectral domains, enhanc- ing the extraction and utilization of dynamic MRI data. The network also introduces a task interaction module that es- tablishes higher-order consistency between segmentation and regression, thus fostering inter-task synergy and improving overall performance. Additionally, we designed a novel task- driven discriminator (TDD) to capture internal high-order re- lationships between tasks. For dynamic MRI information ex- traction, we employ a shallow Transformer network to per- form positional encoding, which captures the relationships within dynamic MRI sequences. In experiments on a dataset of 238 subjects, MTI-Net demonstrates high performance across multiple tasks, indicating its strong potential for as- sisting in the clinical assessment of liver tumors. The code is available at: https://github.com/xiaojiao929/MTI-Net . Introduction Liver cancer is the second leading cause of cancer-related deaths globally (Tan et al. 2024). The segmentation, dy- namic enhancement regression, and classification of liver tu- mors are clinically significant tasks for diagnosis (Hwang et al. 1997; Seo et al. 2019; Zhao et al. 2020; Xiao, Hu, and Wang 2023). For example, as shown in Fig.1(a), the differ- ences in the time-intensity curves between hemangiomas (a benign tumor) and hepatocellular carcinoma (HCC, a malig- nant tumor) provide specific diagnostic insights into these two types of tumors. The clinical value of the dynamic en- hancement process for diagnosing liver tumors is widely recognized (Gupta et al. 2021; Liu et al. 2013). However, Copyright © 2025, Association for the Advancement of Artificial Intelligence (www.aaai.org ). All rights reserved. Figure 1: From left to right, (a) shows the difference of dy- namic enhancement in the time-intensity curve between he- mangioma and HCC. (b) and (c) show the advantages of our method compared to the clinical method. as depicted in Fig.1(c), existing clinical methods still suffer from being labor-intensive, prone to variability, and gener- ally involve multi-step (Xiao et al. 2019). In addition, in- terobserver variability presents another challenge (Kim et al. 2016). Thus, as illustrated in Fig. 1(b), automating and performing the tasks of liver tumor segmentation, dynamic enhancement regression, and classification simultaneously would significantly improve the efficiency of clinical assess- ment and enhance the robustness of diagnosis. Although significant efforts have been made toward au- tomatic liver tumor segmentation and classification (Xiao et al. 2025; Zhao et al. 2020; 2021b), these efforts typi- cally overlook the clinical significance of dynamic enhance- ment curves in distinguishing liver tumors. The simultane- ous multi-task learning of liver tumors remains challenging due to : (1) the absence of an effective end-to-end frame- work to capture the interrelatedness of these tasks for mu- tual improvement, and (2) the lack of a robust mechanism to capture the dependencies across the spatial and tem- poral dimensions of dynamic MRIs for the dynamic en- hancement regression: T1 non-contrast enhanced MRI (Pre- phase), arterial phase CEMRI (Art-phase), portal-venous phase CEMRI (PV-phase), and delay phase CEMRI (Delay- phase). Although traditional convolutional neural network (CNN)-based frameworks excel in local feature extraction, they are limited in capturing global dependencies (Jader- berg et al. 2015; Wang et al. 2018), (e.g., long-range depen- arXiv:2511.20793v1 [eess.IV] 25 Nov 2025 dencies in dynamic MRIs discussed here). Moreover, these frameworks often overlook the inherent periodic patterns and regular changes in signal intensity associated with dy- namic contrast enhancement. In this study, we develop a novel Multi-Task Interac- tive Adversarial Learning Network (MTI-Net) that simul- taneo

This content is AI-processed based on ArXiv data.