Learned end-to-end high-resolution lensless fiber imaging toward intraoperative real-time cancer diagnosis

TL;DR

This paper introduces an end-to-end lensless fiber imaging system that enhances resolution and enables real-time tumor diagnosis during surgery, overcoming limitations of traditional fiber bundle imaging.

Contribution

The study presents a novel resolution enhancement and classification framework that improves image quality and tumor recognition accuracy in lensless fiber endoscopy.

Findings

Resolution enhancement improves image detail beyond physical limits.

Tumor classification accuracy increases from 90.8% to 95.6%.

Framework enables real-time, minimally invasive intraoperative diagnosis.

Abstract

Endomicroscopy is indispensable for minimally invasive diagnostics in clinical practice. For optical keyhole monitoring of surgical interventions, high-resolution fiber endoscopic imaging is considered to be very promising, especially in combination with label-free imaging techniques to realize in vivo diagnosis. However, the inherent honeycomb-artifacts of coherent fiber bundles (CFB) reduce the resolution and limit the clinical applications. We propose an end-to-end lensless fiber imaging scheme toward intraoperative real-time cancer diagnosis. The framework includes resolution enhancement and classification networks that use single-shot fiber bundle images to provide both high-resolution images and tumor diagnosis result. The well-trained resolution enhancement network not only recovers high-resolution features beyond the physical limitations of CFB, but also helps improving tumor recognition rate. Especially for glioblastoma, the resolution enhancement network helps increasing the classification accuracy from 90.8% to 95.6%. The novel technique can enable histological real-time imaging through lensless fiber endoscopy and is promising for rapid and minimal-invasive intraoperative diagnosis in clinics.