New paper: unsupervised learning for medical images analysis

We recently published a new paper on Journal of Supercomputing. The article focuses on an approach that combines automatic feature extraction and unsupervised learning for the analysis of medical images. The idea is to extract (long) vectors of Haralick features for each pixel, and use that information to train a Self-Organizing Map: similar pixels will be clustered together, while anomalous areas (e.g., boundaries or tumors in MRI images) are detected and clustered separately. Both feature extraction and SOM training are offloaded to the GPU using CUDA, which motivates the article.