Pathological examination is essential for cancer diagnosis, with frozen sectioning traditionally serving as the gold standard for intraoperative tissue assessment. However, this method is hindered by complex processing steps and often produces suboptimal tissue slide quality. To address these challenges, a deep-learning-assisted ultraviolet light-emitting diode (UV-LED) microscope has been developed for label-free and slide-free tissue imaging, targeting UV-based light-sheet (UV-LS) images.

A dual-modality imaging system has been constructed to acquire source domain data from UV-LED and target domain data from UV-LS, minimizing tissue and image distortion for training the deep-learning network. The source domain, represented by UV-LED images, captures the projection of the autofluorescence signal within the UV penetration depth, while the target domain, represented by UV-LS images, is one of the 2D image layers (optical sectioning thickness = 1.8 μm) selected from a 3D stack within the UV penetration depth. Consequently, the two image domains are not perfectly paired but share a high degree of morphological similarity. A unified deep-learning framework, termed U-Frame, has been applied to automatically adjust tolerance size according to the image misalignment, effectively handling partially paired data without necessitating exact registration between image pairs. Additionally, the potential of virtual H&E staining on CE-LED images using the U-Frame network is demonstrated to further enhance interpretability for clinical use.

This method achieves an image acquisition speed of 47 seconds/cm² for the contrast-enhanced UV-LED images, approximately 25 times faster than the UV-LS system. The results demonstrate that this approach significantly improves the image quality of UV-LED, revealing critical tissue structures in cancerous samples and published entitled "High-throughput, nondestructive, and low-cost histological imaging with deep-learning-assisted UV microscopy" in Advanced Imaging.

Consequently, the contrast-enhanced UV-LED (CE-LED) offers a low-cost, nondestructive, and high-throughput alternative for histological imaging in intraoperative cancer detection.

Dual-modality tissue imaging system for UV-LED and UV-LS image acquisition with the deep-learning-based contrast enhancement framework