Three-Dimensional Siamese Multi-Level Features Neural Network Based 3D Fusion Improves the Depth of Field in...

Area of Science:

• Biomedical Imaging
• Optical Microscopy
• Photoacoustics

Background:

• Microscopic imaging faces challenges in achieving high resolution and large depth of field (DoF) due to hardware limitations, particularly objective lens focusing.
• Optical-resolution photoacoustic microscopy (OR-PAM) is constrained by a narrow DoF, stemming from the intense laser focusing required for high-resolution imaging.

Purpose of the Study:

• To introduce a novel, cost-effective volumetric information fusion method for achieving large-DoF imaging in OR-PAM.
• To address the inherent DoF limitations of OR-PAM using advanced deep learning techniques.

Main Methods:

• A three-dimensional siamese multi-level features convolutional neural network (3DSMFCNN) was employed for volumetric information fusion.
• Focus region identification was performed on multi-focus 3D photoacoustic data to generate an initial decision map (IDM).
• The IDM was refined using consistency verification and Gaussian filtering to create a final decision map (FDM), enabling voxel-weighted averaging for DoF enhancement.

Main Results:

• The proposed 3DSMFCNN method successfully extended the DoF of OR-PAM imaging.
• Experiments demonstrated that the enhanced DoF did not compromise the lateral resolution of the images.
• The method proved effective, robust, and applicable across simulated and real 3D photoacoustic data, including fibers and blood vessels.

Conclusions:

• The novel 3D deep learning approach significantly enhances the depth of field in OR-PAM.
• This technique offers a cost-effective solution for large-DoF imaging without sacrificing image resolution.
• The findings confirm the method's effectiveness and applicability in advanced biomedical imaging scenarios.