Brain Tumor Anomaly Detection via Latent Regularized Adversarial Network

Nan Wang,Chengwei Chen,Yuan Xie,Lizhuang Ma

DOI: https://doi.org/10.48550/arXiv.2007.04734

2020-07-09

Abstract:With the development of medical imaging technology, medical images have become an important basis for doctors to diagnose patients. The brain structure in the collected data is complicated, thence, doctors are required to spend plentiful energy when diagnosing brain abnormalities. Aiming at the imbalance of brain tumor data and the rare amount of labeled data, we propose an innovative brain tumor abnormality detection algorithm. The semi-supervised anomaly detection model is proposed in which only healthy (normal) brain images are trained. Model capture the common pattern of the normal images in the training process and detect anomalies based on the reconstruction error of latent space. Furthermore, the method first uses singular value to constrain the latent space and jointly optimizes the image space through multiple loss functions, which make normal samples and abnormal samples more separable in the feature-level. This paper utilizes BraTS, HCP, MNIST, and CIFAR-10 datasets to comprehensively evaluate the effectiveness and practicability. Extensive experiments on intra- and cross-dataset tests prove that our semi-supervised method achieves outperforms or comparable results to state-of-the-art supervised techniques.

Image and Video Processing,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are data imbalance and scarce labeled data in brain tumor anomaly detection. Specifically, due to the significant difference in the number of normal and abnormal samples in the brain tumor dataset, and the difficulty in obtaining and labeling a large number of abnormal samples (i.e., brain tumor images), this poses challenges to existing supervised learning methods. These problems result in limited robustness and generalization ability of existing methods when dealing with different types of brain tumors.

To solve the above problems, the paper proposes an innovative semi - supervised anomaly detection algorithm, which is trained only using healthy (normal) brain images. By capturing the common patterns of normal images and detecting anomalies based on the reconstruction error in the latent space, this method can make normal and abnormal samples more distinguishable at the feature level. In addition, this method uses singular values to constrain the latent space for the first time and jointly optimizes the image space through multiple loss functions, thereby improving the detection performance of the model.

Overall, this research aims to improve the accuracy and robustness of brain tumor anomaly detection through a new semi - supervised learning framework, especially in the case of data imbalance and scarce labeled data.