Multi-focus Image Fusion for Confocal Microscopy Using U-Net Regression Map.

Area of Science:

• Biomedical Imaging
• Deep Learning
• Vascular Biology

Background:

• Understanding the spatial relationship between blood and lymphatic vessels in the dura mater is crucial for modeling fluid dynamics in disease.
• Current imaging methods may not fully capture the intricate vascular structures.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based approach for fusing multi-channel microscopy images of the mice dura mater.
• To accurately capture both blood and lymphatic vascular structures in a single fused image.

Main Methods:

• A deep architecture named Multi-Channel Fusion U-Net (MCFU-Net) was developed.
• MCFU-Net fuses multi-channel, single-focus microscopy images within z-stacks.
• It utilizes max pooling for channel-independent estimation of slice-based focus selection maps.

Main Results:

• MCFU-Net accurately captures vascular structures in the mice dura mater.
• The proposed method outperforms the multi-scale Hessian fusion method, avoiding artifacts like dark-halos and non-homogeneous backgrounds.
• Image quality assessment metrics (PIQE, NIQE, BRISQUE) confirmed superior performance.

Conclusions:

• MCFU-Net provides an effective solution for enhanced visualization of vascular networks in the dura mater.
• The method improves anatomical detail and image quality, aiding in disease process modeling.