DTA

Digital Theses Archive

 

Tesi etd-05042018-174939

Type of thesis
Perfezionamento
Author
DI LASCIO, NICOLE
URN
etd-05042018-174939
Title
A machine learning system for carotid plaque vulnerability assessment based on ultrasound images
Scientific disciplinary sector
MED/11
Course
SCIENZE MEDICHE - Translational Medicine
Committee
relatore LIONETTI, VINCENZO
Membro Dott.ssa KUSMIC, CHIARA
Membro Prof.ssa BOLLINI, SVEVA
Membro Prof. GHIADONI, LORENZO
Keywords
  • atherosclerosis
  • machine learning
  • mouse models
  • plaque vulnerability
Exam session start date
;
Availability
parziale
Abstract
This Ph.D project was focussed on the study of carotid atherosclerosis and on the development of an innovative system, based on the elaboration of US images only, aimed at assessing plaque vulnerability. To this purpose, at the beginning of the three-years period related to the Ph.D project, a specific mouse model of carotid atherosclerosis was set up. In particular, ApoE-/- mice were treated with high-fat diet and underwent the surgical procedure for the placement of a perivascular shear stress modifier with a cone-shaped lumen. This particular cast induces the formation of two altered shear stress regions, a low shear stress region in the upstream zone and an oscillatory shear stress region in the downstream one. These two alterations cause the formation of two different atherosclerotic plaques at both the sides of the cast: a vulnerable lesion in the upstream zone and a stable one in the downstream region (Cheng et al., Circulation 2006). Therefore, using this particular murine model, it was possible to study both the plaque phenotypes in each animal, thus reducing the total variability. All the animals included in the experimental protocol were examined using a high-frequency ultrasound system before the surgical procedure and after 3 months from it. At both the time points, ultrasound images were acquired with different modalities (B-mode, Pulsed-Wave Doppler, Color-Doppler) in correspondence of both the sides of the cast and properly elaborated in order to assess morphological, functional and hemodynamic information characterizing the two atherosclerotic plaques. All these parameters were then combined in order to create a mathematical predictive model able to discriminate between stable and unstable lesion. To this purpose different machine learning approaches were tested and the final one was chosen as that providing the best diagnostic performances. The results obtained by means of the ultrasound examinations were compared with immune-histological data achieved performing standard staining.<br>
Files