Abstract

Registering images from different modalities is an active area of research incomputer aided medical interventions. Several registration algorithms have beendeveloped, many of which achieve high accuracy. However, these results aredependent on many factors, including the quality of the extracted features orsegmentations being registered as well as the initial alignment. Althoughseveral methods have been developed towards improving segmentation algorithmsand automating the segmentation process, few automatic initializationalgorithms have been explored. In many cases, the initial alignment from whicha registration is initiated is performed manually, which interferes with theclinical workflow. Our aim is to use scene classification in endoscopicprocedures to achieve coarse alignment of the endoscope and a preoperativeimage of the anatomy. In this paper, we show using simulated scenes that aneural network can predict the region of anatomy (with respect to apreoperative image) that the endoscope is located in by observing a singleendoscopic video frame. With limited training and without any hyperparametertuning, our method achieves an accuracy of 76.53 (+/-1.19)%. There are severalavenues for improvement, making this a promising direction of research. Code isavailable at https://github.com/AyushiSinha/AutoInitialization.

 

Introduction (beta)

None

 

Conclusion (beta)

None