Community Detection in Medical Image Datasets: Using Wavelets and Spectral Methods

Roozbeh Yousefzadeh

Abstract: Medical image datasets may contain a large number of images representing patients with different health conditions. When dealing with raw unlabeled datasets, the large number of samples often makes it hard for experts and non-experts to understand the variety of images present in a dataset. Here, we propose an algorithm to facilitate the automatic identification of communities in medical image datasets. We further demonstrate that such analysis can be insightful in a supervised setting when the images are already labeled. Such insights are useful because health and disease severity can be considered a continuous spectrum. In our approach, we use wavelet decomposition of images in tandem with spectral methods. We show that the eigenvalues of a graph Laplacian can reveal the number of notable communities in an image dataset. Moreover, analyzing the similarities may be used to infer a spectrum representing the severity of the disease.