AttCo: Attention-based co-Learning fusion of deep feature representation for medical image segmentation using multimodality

Duy-Phuong Dao ¹, Hyung-Jeong Yang ¹, Soo-Hyung Kim ¹

⁰Department of Artificial Intelligence Convergence, Chonnam National University, Gwangju, South Korea.

Neural Networks : the Official Journal of the International Neural Network Society +

|

January 14, 2026

View abstract on PubMed

Summary

AttCo, a new multimodal network, improves 3D medical image segmentation by fusing features from different imaging types. This attention-based approach enhances accuracy in identifying abnormal tissues for better healthcare outcomes.

Area Of Science

Medical Imaging
Computer Vision
Artificial Intelligence

Background

Accurate tissue segmentation is vital for disease prediction and treatment planning.
Current 3D medical image segmentation methods struggle with multimodal data fusion and complex structures.

Purpose Of The Study

To introduce AttCo, a novel multimodal semantic segmentation network.
To enhance the learning of complementary information from multiple imaging modalities for 3D segmentation.

Main Methods

Developed AttCo, an attention-based co-learning fusion network for multimodal 3D semantic segmentation.
Employed multiple encoder branches for unimodal 3D representation extraction.
Integrated intra-modality (SEAT) and inter-modality (OSCAT) feature learning for comprehensive fusion.

Main Results

AttCo significantly outperforms existing methods in Dice score across various datasets.
The network effectively extracts robust unimodal 3D representations.
Demonstrated effective exploitation of inter- and intra-modality feature interactions.

Conclusions

AttCo offers a superior approach to multimodal 3D medical image segmentation.
The proposed attention-based fusion effectively captures complex feature interactions.
This method advances clinical analysis through precise abnormal tissue identification.

Keywords:

Attention Co-learning Deep learning Medical imaging segmentation Multimodal segmentation