Abstract:One primary technical challenge in photoacoustic microscopy (PAM) is the necessary compromise between spatial resolution and imaging speed. In this study, we propose a novel application of deep learning principles to reconstruct undersampled PAM images and transcend the trade-off between spatial resolution and imaging speed. We compared various convolutional neural network (CNN) architectures, and selected a fully dense U-net (FD U-net) model that produced the best results. To mimic various undersampling conditions in practice, we artificially downsampled fully-sampled PAM images of mouse brain vasculature at different ratios. This allowed us to not only definitively establish the ground truth, but also train and test our deep learning model at various imaging conditions. Our results and numerical analysis have collectively demonstrated the robust performance of our model to reconstruct PAM images with as few as 2% of the original pixels, which may effectively shorten the imaging time without substantially sacrificing the image quality.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the inherent trade - off between imaging speed and spatial resolution in Photoacoustic Microscopy (PAM). In traditional PAM systems, in order to ensure high spatial resolution, a small step size is required for raster scanning, which results in a slow imaging speed. If the scanning step size is increased to improve the imaging speed, the spatial resolution and image quality will be sacrificed. The paper proposes a deep - learning - based method, especially using the Fully Dense U - net (FD U - net) model, to reconstruct undersampled PAM images, thereby significantly reducing the imaging time without significantly degrading the image quality. Specifically, the paper solves this problem through the following steps: 1. **Data Preparation**: A large number of PAM images of living mouse cerebrovascular were collected and pre - processed, including denoising, contrast enhancement, and grayscale intensity rescaling. 2. **Model Selection and Training**: Multiple Convolutional Neural Networks (CNN) architectures were compared, and finally the FD U - net model with the best performance was selected. This model was trained by the supervised learning method, using Mean Absolute Error (MAE) and Fourier Mean Absolute Error (FMAE) as loss functions to minimize the difference between the reconstructed image and the original fully - sampled image. 3. **Down - sampling and Up - sampling**: In order to simulate different undersampling conditions in practice, the researchers artificially down - sampled the fully - sampled PAM images and used the zero - padding method to restore the missing pixels. Then, the trained FD U - net model was used to perform up - sampling reconstruction on these undersampled images. 4. **Performance Evaluation**: Through indicators such as Peak Signal - to - Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Mean Squared Error (MSE), the reconstruction performance of the model under different down - sampling ratios was evaluated and compared with the traditional Bicubic Interpolation method. The results of the paper show that even when only 2% of the original pixels are used, the FD U - net model can still effectively reconstruct PAM images, significantly improving the imaging speed while maintaining high image quality. This method provides a software - level solution for the development of PAM technology, avoiding expensive hardware improvements.

Reconstructing undersampled photoacoustic microscopy images using deep learning

Deep image prior for undersampling high-speed photoacoustic microscopy

Imaging Enhancement for Acoustic-Resolution Photoacoustic Microscopy by Deep Learning

Photoacoustic microscopy with sparse data by convolutional neural networks

Photoacoustic Microscopy with Sparse Data Enabled by Convolutional Neural Networks for Fast Imaging

UPAMNet: A unified network with deep knowledge priors for photoacoustic microscopy

De-Noising of Photoacoustic Microscopy Images by Deep Learning

Deep-Learning Image Reconstruction for Real-Time Photoacoustic System

Unsupervised Learning for Enhanced Computed Photoacoustic Microscopy

Reconstructing Cancellous Bone from Down-Sampled Optical-Resolution Photoacoustic Microscopy Images with Deep Learning

Accelerating photoacoustic microscopy by reconstructing undersampled images using diffusion models

Deep learning-based photoacoustic imaging of vascular network through thick porous media

Resolution Enhancement of Under-sampled Photoacoustic Microscopy Images using Implicit Neural Representations

Photoacoustic Microscopy Imaging from Acoustic Resolution to Optical Resolution Enhancement with Deep Learning

Reconstructing the Photoacoustic Image with High Quality using the Deep Neural Network Model

High-fidelity deconvolution for acoustic-resolution photoacoustic microscopy enabled by convolutional neural networks

HSD-PAM: High Speed Super Resolution Deep Penetration Photoacoustic Microscopy Imaging Boosted by Dual Branch Fusion Network

Denoising Method for Photoacoustic Microscopy Using Deep Learning

Deep Learning-Based Optical-Resolution Photoacoustic Microscopy for In Vivo 3D Microvasculature Imaging and Segmentation

HDN:Hybrid Deep-learning and Non-line-of-sight Reconstruction Framework for Photoacoustic Brain Imaging

Deep Learning-Powered Biomedical Photoacoustic Imaging