Brain Morphometry by Probabilistic Latent Semantic Analysis

The paper proposes a new shape morphometry approach to combine advanced classification techniques with geometric features in order to identify morphological abnormalities on brain surface. The overall aim is to improve the classification accuracy in distinguishing between normal subjects and patients affected by Schizophrenia. Being inspired by the approaches for natural language processing, local surface geometric patterns are quantized to visual words, and their co-occurrences are encoded as visual topic. To this aim, a generative model is estimated for each of the two populations by employing the probabilistic Latent Semantic Analysis (pLSA) on shapes. Finally, the generative scores observed for each subject are the input of a Support Vector Machine (SVM), which is properly designed to implement a generative-discriminative classification paradigm. An exhaustive experimental section is proposed on a dataset consisting of MRI scans from 64 patients and 60 control subjects. Promisi...

• Type: Journal
• Year: 2010
• Where: MICCAI
• Authors: Umberto Castellani, Alessandro Perina, Vittorio Murino, M. Bellani, G. Rambaldelli, M. Tansella, P. Brambilla
• Keywords: Medical Imaging, MICCAI 2010, Probabilistic Latent Semantic Analysis, Shape Morphometry Approach, Surface Geometric Patterns