Tissue segmentation in volumetric laser endomicroscopy data using U-net and a domain-specific loss function

Joost van der Putten, Fons van der Sommen, Maarten Struyvenberg, Jeroen de Groof, Wouter Curvers, Erik Schoon, Jaques Bergman, Peter H.N. de With

Abstract: Volumetric Laser Endomicroscopy (VLE) is a promising balloon based imaging technique for detecting early neoplasia in Barrett's Esophagus. Especially Computer Aided Detection (CAD) techniques show great promise compared to doctors, who cannot reliably find disease patterns in the VLE signal. However, the relevant tissue has to be segmented in order for these systems to function properly. At present, tissue segmentation has to be done manually and is therefore not scalable for full VLE scans of 1,200 x 4,096 x 2,048 pixels. Furthermore, the current CAD methods cannot use the VLE scans to their full potential as only a small section is selected while an automated system can delineate the entire image. This paper explores the possibility of automatically segmenting relevant tissue for VLE scans using a convolutional neural network. The contribution of this work is threefold. First, this is the first tissue segmentation algorithm for VLE scans. Second, we introduce a weighted ground truth that exploits the signal to noise ratio characteristics of the data. Third, we compare our algorithm segmentations against two additional VLE experts. The results show that our approach is on par with the experts and can therefor be used as a preprocessing step for further classification of the tissue.

Keywords: Volumetric laser endomicroscopy, Deep learning, U-net

Author Affiliation: Eindhoven University of Technology, Academic medical center Amsterdam, Catharina Hospital Eindhoven