Self-supervised Learning for Dense Depth Estimation in Monocular Endoscopy

TL;DR

This paper introduces a self-supervised deep learning method for dense depth estimation in monocular endoscopy that leverages sequential video data and multi-view stereo reconstruction, eliminating the need for manual labeling or prior anatomical models.

Contribution

It presents a novel self-supervised approach that trains CNNs for depth estimation using only endoscopic videos and stereo reconstruction, without manual annotations or prior anatomical knowledge.

Findings

Achieved submillimeter residual errors in depth prediction

Validated method on sinus endoscopy data from two patients

No manual interaction or CT required during training and application

Abstract

We present a self-supervised approach to training convolutional neural networks for dense depth estimation from monocular endoscopy data without a priori modeling of anatomy or shading. Our method only requires sequential data from monocular endoscopic videos and a multi-view stereo reconstruction method, e.g. structure from motion, that supervises learning in a sparse but accurate manner. Consequently, our method requires neither manual interaction, such as scaling or labeling, nor patient CT in the training and application phases. We demonstrate the performance of our method on sinus endoscopy data from two patients and validate depth prediction quantitatively using corresponding patient CT scans where we found submillimeter residual errors.