Extended depth-of-field microscopic imaging for intact histological samples via self-calibrated diplane network

Pathological imaging of surgically excised specimens is the gold standard in clinical practice for diagnosing diseases and making treatment decisions. However, the limited depth of field (DOF) of conventional optical microscopy requires fixing and sectioning of excised samples, which is a cumbersome and lengthy process that significantly increases the time cost of acquiring pathological images. In this study, we propose deep-learning-powered, extended-DOF, dark-field reflectance ultraviolet microscopy (DE-DRUM) for rapid and large-DOF imaging of surgically resected tissues. The system first integrates two cameras positioned at two distinct imaging planes to simultaneously acquire images and then combines the images with a self-calibrated diplane network. The overall DOF of the system is enhanced over eight-folds, and the image acquisition speed is increased by more than seven times. By incorporating the diplane network into a custom-made vacuum compression device, intact and unprocessed mouse brain and human glioma samples are imaged. The results show that DE-DRUM captures pathological features across highly uneven tissue surfaces. DE-DRUM is a rapid, large-DOF, and label-free histological imaging method that holds great potential for application in clinical histopathological diagnosis. In addition, the design of the diplane imaging setup and network provides a plug-and-play solution for large-DOF imaging of various wide-field imaging modalities.