DLLP: a deep learning-based layer prediction network for three-dimensional fluorescence microscopy.

Area of Science:

• Biomedical Imaging
• Microscopy
• Computational Biology

Background:

• High light-throughput microscopy is crucial for biological research, but achieving high speed and throughput simultaneously is challenging.
• Current methods often require additional hardware or compromise image quality.

Purpose of the Study:

• To develop an advanced framework for accelerating 3D microscopy.
• To improve imaging speed and quality without additional optical hardware.

Main Methods:

• Introduced the deep learning-based layer predictor (DLLP) framework.
• Integrated a convolutional neural network (CNN) with an inter-layer dynamic and morphological attention mechanism (IDMA) in a transformer architecture.
• Employed tomographic prediction, unsupervised denoising, and sparse Z-axis recovery.

Main Results:

• Reduced the number of scanning layers in 3D microscopy by over 70% while maintaining light throughput and image fidelity.
• Significantly improved imaging speed and quality through denoising and sparse recovery.
• Demonstrated consistent performance across STED, FMOST, multi-photon, and light-sheet microscopy.

Conclusions:

• DLLP offers a powerful solution for high-throughput, high-speed 3D microscopy.
• The framework surpasses traditional methods and existing deep learning approaches in accuracy and image quality.
• DLLP has broad applicability across various advanced microscopy techniques.