A Mutual Inclusion Mechanism for Precise Boundary Segmentation in Medical Images

TL;DR

MIPC-Net addresses the challenge of precise boundary segmentation in medical images by introducing a Mutual Inclusion of Position and Channel Attention (MIPC) that tightly couples channel and spatial features. The architecture combines CNN and Transformer encoders with the MIPC-Block and GL-MIPC-Skip-Connections to preserve and restore boundary details during decoding. Ablation and cross-dataset experiments show consistent gains in Dice and Hausdorff Distance, with notable 2.23 mm HD improvement on Synapse, signaling strong boundary delineation capabilities. The approach demonstrates robust generalization across CT, dermoscopy, and microscopy data, offering a practical tool for improved disease quantification and treatment planning.

Abstract

In medical imaging, accurate image segmentation is crucial for quantifying diseases, assessing prognosis, and evaluating treatment outcomes. However, existing methods lack an in-depth integration of global and local features, failing to pay special attention to abnormal regions and boundary details in medical images. To this end, we present a novel deep learning-based approach, MIPC-Net, for precise boundary segmentation in medical images. Our approach, inspired by radiologists' working patterns, features two distinct modules: (i) \textbf{Mutual Inclusion of Position and Channel Attention (MIPC) module}: To enhance the precision of boundary segmentation in medical images, we introduce the MIPC module, which enhances the focus on channel information when extracting position features and vice versa; (ii) \textbf{GL-MIPC-Residue}: To improve the restoration of medical images, we propose the GL-MIPC-Residue, a global residual connection that enhances the integration of the encoder and decoder by filtering out invalid information and restoring the most effective information lost during the feature extraction process. We evaluate the performance of the proposed model using metrics such as Dice coefficient (DSC) and Hausdorff Distance (HD) on three publicly accessible datasets: Synapse, ISIC2018-Task, and Segpc. Our ablation study shows that each module contributes to improving the quality of segmentation results. Furthermore, with the assistance of both modules, our approach outperforms state-of-the-art methods across all metrics on the benchmark datasets, notably achieving a 2.23mm reduction in HD on the Synapse dataset, strongly evidencing our model's enhanced capability for precise image boundary segmentation. Codes will be available at https://github.com/SUN-1024/MIPC-Net.

Figures (9)

• Figure 1: Comparison of attention mechanisms used in different medical image segmentation models: (a) only attention, (b) only channel or position attention, (c) integration of position and channel attention, and (d) Mutual nclusion of position and channel attention proposed in this work, which enhances the focus on channel information when extracting position features and vice versa"
• Figure 2: The illustration of the proposed MIPC-Net is depicted. For input medical images, they are fed into the encoder equipped with Transformer mechanisms and MIPC-Block. Subsequently, the features are restored to the original feature maps through the GL-MIPC-Skip-Connections and the decoder. This process yields the final image prediction results.
• Figure 3: The proposed Position and Channel Mutual Inclusion Block (MIPC-Block) integrates positional, channel, and residual mechanisms. In Part A, attention is directed towards channels during the extraction of positional features, while in Part C, the reverse is applied.
• Figure 4: The specific structure of the last Residual module in MIPC-Block.
• Figure 5: Architecture of Dual Attention Block (DA-Block).