Brain Lesion Synthesis via Progressive Adversarial Variational Auto-Encoder

TL;DR

This paper introduces PAVAE, a progressive adversarial variational auto-encoder framework that synthesizes realistic brain lesions to augment training data for improved ROI segmentation in LITT treatments.

Contribution

The paper proposes a novel two-stage lesion synthesis framework with condition and mask embedding blocks to generate diverse, realistic brain lesions for data augmentation.

Findings

Synthetic lesions improve segmentation accuracy.

PAVAE outperforms traditional data augmentation methods.

Realistic lesion synthesis enhances deep learning model training.

Abstract

Laser interstitial thermal therapy (LITT) is a novel minimally invasive treatment that is used to ablate intracranial structures to treat mesial temporal lobe epilepsy (MTLE). Region of interest (ROI) segmentation before and after LITT would enable automated lesion quantification to objectively assess treatment efficacy. Deep learning techniques, such as convolutional neural networks (CNNs) are state-of-the-art solutions for ROI segmentation, but require large amounts of annotated data during the training. However, collecting large datasets from emerging treatments such as LITT is impractical. In this paper, we propose a progressive brain lesion synthesis framework (PAVAE) to expand both the quantity and diversity of the training dataset. Concretely, our framework consists of two sequential networks: a mask synthesis network and a mask-guided lesion synthesis network. To better employ extrinsic information to provide additional supervision during network training, we design a condition embedding block (CEB) and a mask embedding block (MEB) to encode inherent conditions of masks to the feature space. Finally, a segmentation network is trained using raw and synthetic lesion images to evaluate the effectiveness of the proposed framework. Experimental results show that our method can achieve realistic synthetic results and boost the performance of down-stream segmentation tasks above traditional data augmentation techniques.