ResDAC-Net: a novel pancreas segmentation model utilizing residual double asymmetric spatial kernels

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Deep Learning

Background:

• Pancreatic segmentation is challenging due to complex abdominal anatomy and blurred organ boundaries.
• Accurate pancreatic segmentation is vital for computer-aided diagnosis, surgical planning, and organ assessment.

Purpose of the Study:

• To propose a novel Residual Double Asymmetric Convolution Network (ResDAC-Net) for accurate pancreatic tissue segmentation.
• To enhance the performance of computer-aided diagnosis systems through improved pancreatic segmentation.

Main Methods:

• Development of a novel ResDAC-Net model incorporating specialized ResDAC blocks.
• Implementation of feature fusion between adjacent encoding layers to utilize multi-level features.
• Application of parallel dilated convolutions to capture multiscale spatial information and increase receptive field.

Main Results:

• The ResDAC-Net model effectively highlights pancreatic features and captures multiscale spatial information.
• Feature fusion strategy leverages both low-level and deep-level features for enhanced segmentation.
• The model demonstrates high compatibility with state-of-the-art models, excelling in key segmentation metrics like DSC and Jaccard index.

Conclusions:

• ResDAC-Net offers a significant advancement in pancreatic segmentation accuracy.
• The proposed model contributes to more reliable computer-aided diagnosis systems for pancreatic conditions.
• ResDAC-Net shows strong potential for clinical applications in surgical planning and organ assessment.