MCA-UNet

MCA-UNet segments medical images by integrating multi-scale representations, dense skip connections, and cross co-attention into a U-Net architecture to improve delineation of pathological structures such as consolidation, ground-glass opacity (GGO), microaneurysms (MA), and hard exudates (EX).

Key Features:

• Multi-Scale Feature Modeling: constructs detailed multi-scale representations and employs dense skip connections to reduce semantic gaps and capture global contextual information.
• Cross Co-Attention Mechanism: jointly models spatial and channel attentions via cross co-attention to enhance focus on relevant features across different scales.
• Improved Segmentation Accuracy: achieves superior segmentation performance on datasets such as COVID-19 and IDRiD for delineating consolidation, GGO, MA, and EX.

Scientific Applications:

• COVID-19 Imaging: enhances detection and segmentation of lung abnormalities including consolidations and ground-glass opacities.
• Diabetic Retinopathy: improves identification and delineation of retinal lesions such as microaneurysms and hard exudates.

Methodology:

The method integrates dense skip connections with cross co-attentional mechanisms within the U-Net framework to mitigate semantic gaps and capture both local and global features at multiple scales.