Advanced kidney mass segmentation using VHUCS-Net with protuberance detection network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate segmentation of kidney masses and structures is crucial for diagnosis and treatment.
• Existing methods may struggle with complex structural features and precise boundary identification in kidney images.

Purpose of the Study:

• To introduce the dual-track hybrid VHUCS-Net architecture for enhanced kidney mass and structure segmentation.
• To improve the accuracy and interpretability of segmentation outputs for clinical decision support.

Main Methods:

• Integration of a transformer-enhanced U-Net with a contrast-optimized Protuberance Detection Network (PDN).
• Utilizing Vision Transformer attention and High-Resolution Network (HRNet) for global and high-resolution feature capture.
• Employing multi-scale pooling, contrast enhancement, and feature fusion within the PDN for precise mass segmentation.

Main Results:

• Achieved an Intersection over Union (IoU) score of 0.9441 and a Dice coefficient of 0.9712 on a kidney segmentation dataset.
• Demonstrated outstanding segmentation precision, highlighting structural size-shape variants, boundaries, and complex features.
• Validated generalizability across multiple segmentation tasks using additional public datasets.

Conclusions:

• VHUCS-Net significantly enhances diagnostic efficiency and supports clinical decision-making.
• The model provides accurate, interpretable segmentation outputs for kidney disease analysis.
• The proposed architecture proves effective and generalizable for various medical image segmentation tasks.