One-Class SVM on siamese neural network latent space for Unsupervised Anomaly Detection on brain MRI White Matter Hyperintensities

Nicolas Pinon,Robin Trombetta,Carole Lartizien

2023-04-17

Abstract:Anomaly detection remains a challenging task in neuroimaging when little to no supervision is available and when lesions can be very small or with subtle contrast. Patch-based representation learning has shown powerful representation capacities when applied to industrial or medical imaging and outlier detection methods have been applied successfully to these images. In this work, we propose an unsupervised anomaly detection (UAD) method based on a latent space constructed by a siamese patch-based auto-encoder and perform the outlier detection with a One-Class SVM training paradigm tailored to the lesion detection task in multi-modality neuroimaging. We evaluate performances of this model on a public database, the White Matter Hyperintensities (WMH) challenge and show in par performance with the two best performing state-of-the-art methods reported so far.

Image and Video Processing,Artificial Intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to conduct effective anomaly detection in the absence of supervision information in brain MRI white matter hyperintensities (WMH) detection. Specifically, researchers are concerned with how to use unsupervised methods to detect these anomalies in neuroimaging when the lesions are very small or have weak contrast. This type of problem is particularly important in medical image analysis because pathological images may have heterogeneous patterns and it is very difficult to obtain labels from radiologists. The paper proposes a method based on the latent space constructed by auto - encoders and combined with the one - class support vector machine (One - Class SVM) training paradigm to achieve the lesion detection task in multi - modal neuroimaging.

The main contributions of the paper are as follows:
1. **Propose a new unsupervised anomaly detection (UAD) method**: This method uses siamese auto - encoders to construct the latent space and performs anomaly detection through one - class support vector machines (One - Class SVM).
2. **Evaluate the model performance**: The performance of this model is evaluated on the dataset of the public database WMH challenge, and results comparable to the currently reported best methods are shown.
3. **Improve the robustness of anomaly detection**: By calculating the normal distribution at the individual patient level instead of training on the entire dataset, the detection performance is improved and the dependence on registration quality is reduced.

In summary, the paper aims to develop an unsupervised method that can effectively detect small or low - contrast lesions in brain MRI in the absence of a large amount of labeled data.