Research Interests

MRI-based Quantitative Assessment of Arthritis

Quantitative methods for assessing the activity of a disease and its response to therapies are of extraordinary importance in today's medicine. The main focus of my research in the last years has been the development of MRI-based quantitative methods for assessing juvenile forms of arthritis.

Most of the work has been devoted to the assessment of the synovitis - the inflammation of the synovial membrane - through the analysis of DCE-MRI (see this article) or the automatic measurement of the inflamed synovial volume. On this latter topic we have published at conferences (IEEE ISBI 2009 and ESR 2010), and a journal article has been submitted.

This part of my research has been initially driven by the EU-project Health-e-Child, and by the collaborations born in that context, in particular with Istituto G. Gaslini and its rheumatology and radiology groups. It was joint work with (in strictly alphabetical order): M. B. Damasio, M. Esposito, C. Malattia, A. Martini, M. Santoro, P. Tomà, A. Verri.

Machine Learning Methods for Medical Image Analysis

Example of learned dictionaries

Some of the methods we employ in medical image analysis are based on machine learning. In particular, the measurement of the synovial volume (see above and this paper) relies on a novel voxel classification method that builds a linear classifier out of a large pool of non-linear features. Indeed, as we show in a paper we recently submitted, the resulting classifier is much sparser (and therefore faster) than an SVM, while retaining comparable accuracy.

A second topic of interest is dictionary learning (aka sparse coding). There are comparably few works exploring the advantages of these techniques for medical image analysis, and our group is trying to fill the gap. See this work accepted as poster at MCV 2010, a workshop held in conjuction with MICCAI. At the same time, we are developing new methods for learning dictionaries allowing for fast encoding (analysis) of new data. I recently gave a talk (here are the slides in pdf) on the subject at INSPIRE 2010, and publications are on their way (starting with a short technical report, arXiv:1011.3728v1).

3D Morphable Models (3DMM)

Fig.6 from Basso et al, 2006, JVRB

The application of 3DMM to the analysis and synthesis of facial expressions has been the topic of my PhD (here you can find the thesis). A good place to start if you want to know more on the subject is probably the web page of T.Vetter's group.

If you want to take a look at the work I did, you can check the articles I published on fitting 3DMM using implicit representations and fitting 3DMM to expression data, as well as the one I co-authored on using 3DMM for synthesizing expressions in static images.

This was joint work with (in strictly alphabetical order): V. Blanz, P. Paysan, T. Poggio, S. Romdhani, A. Verri, T. Vetter.