ASC-Net : Adversarial-based Selective Network for Unsupervised Anomaly Segmentation

TL;DR

ASC-Net is an adversarial neural network framework designed for unsupervised anomaly segmentation, effectively partitioning images into normal and anomalous regions without labeled data, and showing significant performance improvements over existing methods.

Contribution

This paper introduces ASC-Net, a novel adversarial-based network that combines clustering and anomaly detection for unsupervised segmentation, leveraging user-defined reference distributions.

Findings

Outperforms existing unsupervised anomaly segmentation methods like AnoGAN

Demonstrates significant improvements on BraTS, LiTS, and MS-SEG2015 datasets

Shows potential for unsupervised learning with user-guided reference distributions

Abstract

We introduce a neural network framework, utilizing adversarial learning to partition an image into two cuts, with one cut falling into a reference distribution provided by the user. This concept tackles the task of unsupervised anomaly segmentation, which has attracted increasing attention in recent years due to their broad applications in tasks with unlabelled data. This Adversarial-based Selective Cutting network (ASC-Net) bridges the two domains of cluster-based deep learning methods and adversarial-based anomaly/novelty detection algorithms. We evaluate this unsupervised learning model on BraTS brain tumor segmentation, LiTS liver lesion segmentation, and MS-SEG2015 segmentation tasks. Compared to existing methods like the AnoGAN family, our model demonstrates tremendous performance gains in unsupervised anomaly segmentation tasks. Although there is still room to further improve performance compared to supervised learning algorithms, the promising experimental results shed light on building an unsupervised learning algorithm using user-defined knowledge.