Whole slide imaging (WSI) is an emerging technology for digital pathology. The accuracy and speed of autofocusing are critical for the performance of the WSI system. This paper introduces a novel technique of deep autofocusing for WSI. Instead of mechanically adjusting the focal distance on a tile-by-tile basis, we develop a deep convolutional neural network for tile-wise autofocusing to generate in-focus images from tentative possibly defocused images. This deep autofocusing network (DAFNet) works with only two images taken at different focal distances; in contrast, traditional methods need to take, for each tile of the target ultra high-resolution pathology image, a stack of as many as 21 shots with varying focal distances. The novel architecture design of DAFNet facilitates the fusion of complementary information of the two input images of different focal distances. The proposed off-line reconstruction strategy allows high throughput scanning of sample slides done without compromising image quality, because DAFNet can rectify errors in focal distance and bring the scanned tiles back into focus by learnt non-linear dual-input blur-to-sharp mapping. Experimental results demonstrate the refocusing capability of the DAFNet method.

An untrained neural network (UNN) paves a new way to realize lensless imaging from single-frame intensity data. Based on the physics engine, such methods utilize the smoothness property of a convolutional kernel and provide an iterative self-supervised learning framework to release the needs of an end-to-end training scheme with a large dataset. However, the intrinsic overfitting problem of UNN is a challenging issue for stable and robust reconstruction. To address it, we model the phase retrieval problem into a dual-constrained untrained network, in which a phase-amplitude alternating optimization framework is designed to split the intensity-to-phase problem into two tasks: phase and amplitude optimization. In the process of phase optimization, we combine a deep image prior with a total variation prior to retrain the loss function for the phase update. In the process of amplitude optimization, a total variation denoising-based Wirtinger gradient descent method is constructed to form an amplitude constraint. Alternative iterations of the two tasks result in high-performance wavefield reconstruction. Experimental results demonstrate the superiority of our method.

Lensless imaging with a mask is an attractive topic as it enables a compact configuration to acquire wavefront information of a sample with computational approaches. Most existing methods choose a customized phase mask for wavefront modulation and then decode the sample’s wave field from modulated diffraction patterns. Different from phase masks, lensless imaging with a binary amplitude mask facilitates a cheaper fabrication cost, but high-quality mask calibration and image reconstruction have not been well resolved. Here we propose a self-calibrated phase retrieval (SCPR) method to realize a joint recovery of a binary mask and sample's wave field for a lensless masked imaging system. Compared with conventional methods, our method shows a high-performance and flexible image recovery without the help of an extra calibration device. Experimental results of different samples demonstrate the superiority of our method.

Single-shot lensless imaging with a binary amplitude mask enables a low-cost and miniaturized configuration for wave field recovery. However, the mask only allows a part of the wave field to be captured, and thus the inverse decoding process becomes a highly ill-posed problem. Here we propose an enhanced self-calibrated phase retrieval (eSCPR) method to realize single-shot joint recovery of mask distribution and the sample’s wavefront. In our method, a sparse regularized phase retrieval (SrPR) algorithm is designed to calibrate the mask distribution. Then, a denoising regularized phase retrieval (DrPR) algorithm is constructed to reconstruct the wavefront of the sample. Compared to conventional single-shot methods, our method shows robust and flexible image recovery. Experimental results of different samples are given to demonstrate the superiority of our method.

Deep learning has emerged as a prominent technique in the field of holographic imaging, owing to its rapidity and high performance. Prevailing deep neural networks employed for holographic image reconstruction predominantly rely on convolutional neural networks (CNNs). While CNNs have yielded impressive results, their intrinsic limitations, characterized by a constrained local receptive field and uniform representation, pose challenges in harnessing spatial texture similarities inherent in holographic images. To address this issue, we propose a novel hierarchical framework based on self-attention mechanism for digital holographic reconstruction, termed HoloFormer. Specifically, we adopt a window-based transformer block as the backbone, significantly reducing computational costs. In the encoder, a pyramid-like hierarchical structure enables the learning of feature map representations at different scales. In the decoder, a dual-branch design ensures that the real and imaginary parts of the complex amplitude do not exhibit cross-talk with each other. During the training phase, we incorporate contrastive regularization to maximize the utilization of mutual information. Overall, our experiments demonstrate that HoloFormer achieves superior reconstruction results compared to previous CNN-based architectures. This progress further propels the development of deep learning-based holographic imaging, particularly in lensless microscopy applications.