How Statistics can improve image processing and analysis

 
Speaker:
Aymeric Stamm e Riccardo Pascuzzo
Affiliation:
Mox - Dipartimento di Matematica, Politecnico di Milano
When:
Friday 19th February 2016
Time:
11:30:00
Where:
Aula Seminari "Saleri" VI Piano MOX-Dipartimento di Matematica, Politecnico di Milano
Abstract:
This brief overview of applications of Statistics to image processing and analysis will be given by two separate speakers. First, Aymeric Stamm will introduce problematics pertaining to magnetic resonance imaging (MRI). A first application in which Statistics will play an important role is diffusion MRI, the goal of which is to provide an assessment of the integrity of the white matter of the Human brain, whose main constituent cell is the axon. The measured diffusion signal reflects how water molecules randomly diffuse within and around the axons. The characterization of such a diffusion process is of primary importance, e.g., for tumor removal surgery, because main directions of diffusion identified in a voxel are putative directions of entire axons passing through that voxel. Many models have been devised to characterize the diffusion process at the voxel level. Much less has been done to design proper estimators of the parameters of such models, together with an appropriate noise model, with subsequent characterization of their statistical properties. A second application in MRI that could use the help of Statistics is image formation. The MRI scanner measures its signal using an array of multiple surface coil elements, each of which capturing different small areas of the brain. Image formation pertains to the combination of coil element images to form a single composite high signal-to-noise and high contrast-to-noise ratio complex image across the whole brain. Next, Riccardo Pascuzzo will present advanced statistical tools for the analysis of digital images. Digital image processing and analysis of information in images are methods that have become increasingly important in many technical and scientific fields. He will focus on statistical methods for modeling (e.g. acquiring, showing, filtering and providing segmentation of images) and classifying images. He will show how these methods have successfully been applied in three different case studies: (i) Horse pain assessment (in collaboration with Dep. of Veterinary Medicine, UNIMI), (ii) Air bubble identification in syringe (project funded by ThermoFisher SpA) and (iii) functional MRI (in collaboration with Ospedale San Raffaele Milano, Unità di Psichiatria e psicobiologia clinica). contact: laura.sangalli@polimi.it
PDF: