DmADs-Net: Dense multiscale attention and depth-supervised network for medical image segmentation

TL;DR

DmADs-Net addresses weak feature localization and edge-detail loss in medical image segmentation by integrating dense multiscale attention (MSCFA), local feature attention (LFA), and a feature refinement and fusion block (FRFB) under a depth-supervised framework using $ResNet_{18}$/$ResNet_{34}$ backbones. The architecture fuses shallow and deep features through MSCFA and FRFB while leveraging LFA to strengthen high-level semantic associations, with deep supervision guiding intermediate reconstructions. Evaluations on five public datasets (JSRT, ISIC2016, DSB2018, BUSI, GlaS) demonstrate superior performance and robust edge/detail handling; ablation studies confirm the importance of each module and the deep supervision strategy. The findings suggest improved localization and edge reconstruction across modalities, supporting potential clinical impact in lesion detection and treatment planning, while outlining future work to incorporate Transformer/diffusion techniques and broaden dataset coverage.

Abstract

Deep learning has made important contributions to the development of medical image segmentation. Convolutional neural networks, as a crucial branch, have attracted strong attention from researchers. Through the tireless efforts of numerous researchers, convolutional neural networks have yielded numerous outstanding algorithms for processing medical images. The ideas and architectures of these algorithms have also provided important inspiration for the development of later technologies.Through extensive experimentation, we have found that currently mainstream deep learning algorithms are not always able to achieve ideal results when processing complex datasets and different types of datasets. These networks still have room for improvement in lesion localization and feature extraction. Therefore, we have created the Dense Multiscale Attention and Depth-Supervised Network (DmADs-Net).We use ResNet for feature extraction at different depths and create a Multi-scale Convolutional Feature Attention Block to improve the network's attention to weak feature information. The Local Feature Attention Block is created to enable enhanced local feature attention for high-level semantic information. In addition, in the feature fusion phase, a Feature Refinement and Fusion Block is created to enhance the fusion of different semantic information.We validated the performance of the network using five datasets of varying sizes and types. Results from comparative experiments show that DmADs-Net outperformed mainstream networks. Ablation experiments further demonstrated the effectiveness of the created modules and the rationality of the network architecture.

Figures (33)

• Figure 1: A comparative experiment diagram of Ours and mainstream networks.We annotate the attention features of the segmentation targets in the images and perform visual overlap processing for all participating methods.Red for incorrectly identified areas, green for unidentified areas.
• Figure 2: DmADs-Net main network diagram. Using $ResNet_{18}$ and $ResNet_{34}$ as the backbone network to complete feature extraction of different depths.In order to preserve more weak feature information, the MSCFA module is created and applied in skip connections and deep networks.LFA is created to enhance the contextual association of high-level semantic information. FRFB is created to complete the fusion of different information. A deep supervision mechanism is used to train the reinforcement network.
• Figure 3: Multi-scale Convolutional Feature Attention Block.
• Figure 4: Feature attention map after MSCFA processing.
• Figure 5: Local Feature Attention Block.