Development of pre and post-processing steps to a pipeline aimed to identify silent cerebral infarcts

Biondini, Nicolas (2023) Development of pre and post-processing steps to a pipeline aimed to identify silent cerebral infarcts. [Laurea magistrale], Università di Bologna, Corso di Studio in Fisica [LM-DM270]
Documenti full-text disponibili:
[thumbnail of Thesis] Documento PDF (Thesis)
Disponibile con Licenza: Creative Commons: Attribuzione - Non commerciale - Condividi allo stesso modo 4.0 (CC BY-NC-SA 4.0)

Download (13MB)

Abstract

Sickle Cell Disease (SCD) is a group of red blood cell disorders which cause abnormal hemoglobin and could lead to different symptoms. Brain MRI scans of patients affected by this condition show peculiar lesions in White Matter, without any apparent neurological evidence, called Silent Cerebral Infarcts (SCI). To increase the comprehension of this condition is necessary to find and segment the lesions. Up to now this process is performed manually, comporting an high consumption of time and a dependence on the experience of the involved operator. The European project GenoMed4All aims to provide solutions, control and pre- vention for haematological diseases, including SCD, by applying AI technologies. In this context, a pipeline for the identification and segmentation of SCI was proposed. This work of thesis aims the to develop and implement a pre-processing and post- processing steps to improve the results obtained with the proposed segmentation technique. The pre-processing step includes a phase of brain automatic extraction and seg- mentation of its main tissues. The post-processing step consists in the classification of the lesions found in the segmentation step, aiming to remove the false positives. The proposed steps were developed and tested on a data set of MRI scans pro- vided by different medical centers in Italy. The performances of the pre-processing step were tested comparing the obtained results with the ones of the software FSL, which is a standard in the analysis of MRI scans. The brain mask comparison mea- sured a Dice coefficient of 0.87, th white matter of 0.78, the grey matter of 0.67 and the cerebrospinal fluid of 0.66. The post-processing step was developed training a set of three classifiers and their results were compared to the manual annotation of the SCIs in the same data set, obtaining a, AUC precision-recall for the best classifier of 0.75.

Abstract
Tipologia del documento
Tesi di laurea (Laurea magistrale)
Autore della tesi
Biondini, Nicolas
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Indirizzo
Curriculum E: Fisica applicata
Ordinamento Cds
DM270
Parole chiave
Brain extraction,skull stripping,mri,brain segmentation,expectation maximization,classifiers,SCD,SCI,Sickle cell disease,Silent cerebral infarct,segmentation
Data di discussione della Tesi
31 Marzo 2023
URI

Altri metadati

Statistica sui download

Gestione del documento: Visualizza il documento

^