Year: 2025
Type: article
Abstract: With the rapid advancement of medical imaging technology, medical image segmentation has become increasingly crucial in disease diagnosis, treatment planning, and intraoperative navigation. In this paper, we propose a novel convolutional module, AdaptiveConv2d, that is designed to address the limitations of traditional convolutions and advance convolutional techniques in medical image processing. This approach targets the unique challenges associated with medical images, including complex tissue structures, irregular boundaries, low contrast, and high noise levels. The AdaptiveConv2d module integrates adaptive feature extraction, an optimized receptive field, enhanced sensitivity to details and boundaries, and an innovative feature fusion mechanism to significantly improve segmentation accuracy and robustness. By dynamically adjusting the convolution operations, the module adapts flexibly to medical images with varying shapes and boundaries, leading to more precise feature extraction. Experimental results indicate that the AdaptiveConv2d module outperforms several existing methods across multiple medical image segmentation tasks, highlighting its potential as an effective tool for medical image analysis. Furthermore, the highly modular design of AdaptiveConv2d allows for seamless integration into existing neural network architectures, offering a versatile and adaptable solution for a range of medical image segmentation applications.
Source: Applied Sciences
DOI: 10.3390/app15042169
Authors: Donghua Liu, Zuofeng Zhou
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
Award: G4494139702
OpenAlex ID: W4407699656