Medical image segmentation network based on a multisize convolutional kernel association strategy.

Background: Medical image segmentation plays a crucial role in clinical diagnosis and treatment planning by extracting essential information from tissue images.

Purpose: This research aims to address limitations in current medical image segmentation models by proposing a new CKASnet model that enhances adaptability and efficiency while maintaining segmentation accuracy.

Methods: The CKASnet model integrates a novel convolutional kernel association strategy (CKAS), which modifies and updates convolutional kernels to improve their receptive fields and adaptability. This approach combines the advantages of transformers' attention mechanisms with convolutional neural networks (CNNs) to better handle complex medical imaging tasks.

Results: Experimental evaluations demonstrate that the CKASnet model outperforms existing segmentation models across multiple datasets. It achieves superior segmentation accuracy by effectively learning intricate features without the need for extensive pretraining.

Conclusions: The CKASnet model represents a significant advancement in medical image segmentation technology, offering enhanced flexibility and performance. Its innovative CKAS demonstrates its potential to improve clinical diagnostic outcomes and pathological analysis.