Deeply Supervised UNet for Semantic Segmentation to Assist Dermatopathological Assessment of Basal Cell Carcinoma.

Jean Le’Clerc Arrastia,Jörg Schaller,Daniel Otero Baguer,Tobias Boskamp,Lena Hauberg-Lotte,Peter Maass,Nicole Duschner,Sonja Hetzer,Nick Heilenkötter

doi:10.3390/jimaging7040071

Abstract

Accurate and fast assessment of resection margins is an essential part of a dermatopathologist’s clinical routine. In this work, we successfully develop a deep learning method to assist the dermatopathologists by marking critical regions that have a high probability of exhibiting pathological features in whole slide images (WSI). We focus on detecting basal cell carcinoma (BCC) through semantic segmentation using several models based on the UNet architecture. The study includes 650 WSI with 3443 tissue sections in total. Two clinical dermatopathologists annotated the data, marking tumor tissues’ exact location on 100 WSI. The rest of the data, with ground-truth sectionwise labels, are used to further validate and test the models. We analyze two different encoders for the first part of the UNet network and two additional training strategies: (a) deep supervision, (b) linear combination of decoder outputs, and obtain some interpretations about what the network’s decoder does in each case. The best model achieves over 96%, accuracy, sensitivity, and specificity on the Test set.

Highlights

Deep learning has shown great potential to address several problems in understanding, reconstructing, and reasoning about images
Convolutional neural network approaches have been actively used for classification and segmentation tasks in a wide field of applications [1,2,3], ranging from robot vision and understanding to the support of critical medical tasks [4,5,6,7]
We aim to develop a deep learning method that can support dermatopathologists in providing fast, reliable, and reproducible decisions for the assessment of basal cell carcinoma (BCC) resection margins

Summary

Introduction

The nearly human-expert performance achieved in some medical imaging applications [8,9] has come to show the capabilities and potential of these algorithms. It enabled the extraction of previously hidden information from routine histology images and a new generation of biomarkers [10]. BCC is the most common malignant skin cancer with an increasing incidence of up to 10% a year [11] It can be locally destructive and is an essential source of morbidity for patients, mainly when located on the face. The resection margins’ microscopic control can reduce the recurrence rate to 1% [14,15]

Objectives

Methods

Results

Discussion

Conclusion