Search for abnormal thermal patterns in clinical thermal infrared imaging

This paper presents a technique to uncover abnormal thermal patterns in clinical thermal infrared images. When little information is known about pathological states, the search for abnormal thermal patterns is difficult. Supervised approaches require extensive knowledge about the distribution of abnormal patterns and are not appropriate for blind searches. We propose an approach that is based on a fusion of clusters from feature images and from an unsupervised clustering adaptive resonance theory (ART) neural network. We show that abnormal thermal patterns can be recovered in a selected number of controlled cases, where the number and location of anomalies are known. First results indicate the potential usefulness of our method for large scale screening of patients.