fourier-glrt-based-graph-classification

Graph-structured data classification based on spectral methods and generalized likelihood-ratio testing

This page documents the reference implementation used in the following work on binary classification of graph-structured data applied to computer-aided diagnosis in neurology.

Blog article

• Title: Graph-structured data classification based on spectral methods and the generalized likelihood ratio test
• Author: Michel Pohl
• Year: 2026
• Link: https://pohl-michel.github.io/blog/articles/fourier-glrt-graph-classification/article.html

Code

• GitHub repository: https://github.com/pohl-michel/fourier-glrt-based-graph-classification

Description

I implemented a binary classifier for graph-structured data, combining generalized likelihood ratio testing and graph Fourier transforms, based on prior work from Hu et al. (2016). I applied this method to Alzheimer's disease detection by modeling brain regions in PET images as nodes in a graph. The edge weights, representing similarity between regional imaging feature values, were computed using a Gaussian RBF kernel. This approach yielded a leave-one-out test F1 score of 0.85 on a dataset of 142 brain scans (61 healthy controls and 81 Alzheimer's disease cases).

Note

The blog article is based on work conducted jointly at Centrale Méditerranée and Fresnel Institute in 2016 under the supervision of Mouloud Adel.