Supervision by Denoising for Medical Image Segmentation

TL;DR

This paper introduces 'supervision by denoising' (SUD), a semi-supervised learning framework that improves medical image reconstruction by using denoised outputs as soft labels, reducing the need for extensive labeled data.

Contribution

The paper proposes SUD, a novel semi-supervised framework that unifies denoising techniques for better medical image reconstruction without handcrafted regularizers.

Findings

SUD significantly improves reconstruction quality over supervised baselines.

Applied to brain reconstruction and cortical parcellation, SUD outperforms existing methods.

Demonstrates effectiveness in domains with scarce labeled data.

Abstract

Learning-based image reconstruction models, such as those based on the U-Net, require a large set of labeled images if good generalization is to be guaranteed. In some imaging domains, however, labeled data with pixel- or voxel-level label accuracy are scarce due to the cost of acquiring them. This problem is exacerbated further in domains like medical imaging, where there is no single ground truth label, resulting in large amounts of repeat variability in the labels. Therefore, training reconstruction networks to generalize better by learning from both labeled and unlabeled examples (called semi-supervised learning) is problem of practical and theoretical interest. However, traditional semi-supervised learning methods for image reconstruction often necessitate handcrafting a differentiable regularizer specific to some given imaging problem, which can be extremely time-consuming. In this work, we propose "supervision by denoising" (SUD), a framework that enables us to supervise reconstruction models using their own denoised output as soft labels. SUD unifies stochastic averaging and spatial denoising techniques under a spatio-temporal denoising framework and alternates denoising and model weight update steps in an optimization framework for semi-supervision. As example applications, we apply SUD to two problems arising from biomedical imaging -- anatomical brain reconstruction (3D) and cortical parcellation (2D) -- to demonstrate a significant improvement in the image reconstructions over supervised-only and stochastic averaging baselines.