Epistemic and Heteroscedastic Uncertainty Estimation in Retinal Blood Vessel Segmentation

Pedro Costa,Aurélio Campilho,Asim Smailagic,Jaime Cardoso

doi:10.24840/2183-6493_007.003_0008

Abstract

Current state-of-the-art medical image segmentation methods require high quality datasets to obtain good performance. However, medical specialists often disagree on diagnosis, hence, datasets contain contradictory annotations. This, in turn, leads to difficulties in the optimization process of Deep Learning models and hinder performance. We propose a method to estimate uncertainty in Convolutional Neural Network (CNN) segmentation models, that makes the training of CNNs more robust to contradictory annotations. In this work, we model two types of uncertainty, heteroscedastic and epistemic, without adding any additional supervisory signal other than the ground-truth segmentation mask. As expected, the uncertainty is higher closer to vessel boundaries, and on top of thinner and less visible vessels where it is more likely for medical specialists to disagree. Therefore, our method is more suitable to learn from datasets created with heterogeneous annotators. We show that there is a correlation between the uncertainty estimated by our method and the disagreement in the segmentation provided by two different medical specialists. Furthermore, by explicitly modeling the uncertainty, the Intersection over Union of the segmentation network improves 5.7 percentage points.

Highlights

Retinal vasculature provides information about many conditions including vision threatening diseases, such as Diabetic Retinopathy, and cardiovascular diseases, such as coronary artery disease (Nguyen and Wong 2009)
Current state-of-the-art blood vessel segmentation methods rely on Convolutional Neural Networks (CNNs) (Imran et al 2019; Meyer et al 2017; Meyer et al 2018) which typically require large high-quality datasets to achieve best performance
The U-Net consists of an encoder-decoder CNN, with skip connections between the encoder and the decoder layers, to Epistemic and Heteroscedastic Uncertainty Estimation in Retinal Blood Vessel Segmentation Pedro Costa, Asim Smailagic, Jaime S

Summary

Introduction

Retinal vasculature provides information about many conditions including vision threatening diseases, such as Diabetic Retinopathy, and cardiovascular diseases, such as coronary artery disease (Nguyen and Wong 2009). The task of segmenting the blood vessels in retinal images is an important first step towards automatically diagnosing these diseases. Current state-of-the-art blood vessel segmentation methods rely on Convolutional Neural Networks (CNNs) (Imran et al 2019; Meyer et al 2017; Meyer et al 2018) which typically require large high-quality datasets to achieve best performance. The best performing methods (Meyer et al 2017; Meyer et al 2018) typically use a U-Net style architecture (Ronneberger, Fischer, and Brox 2015) to segment the input images. The U-Net consists of an encoder-decoder CNN, with skip connections between the encoder and the decoder layers, to Epistemic and Heteroscedastic Uncertainty Estimation in Retinal Blood Vessel Segmentation Pedro Costa, Asim Smailagic, Jaime S. Aurélio Campilho help preserve fine details from the input image in the output segmentation mask, such as the edges of the object of interest

Methods

Results

Conclusion