Early-stage Philadelphia-negative myeloproliferative neoplasms (Ph-MPN) represent a diagnostic challenge due to morphological overlap between polycythemia vera (PV), essential thrombocythemia (ET), and prefibrotic myelofibrosis (pre-PMF). This thesis develops an end-to-end pipeline for automatic classification based exclusively on morphometric analysis of digitized bone marrow biopsies (BMB) using Whole Slide Imaging (WSI). On a retrospective cohort of 101 BMBs (19 pre-PMF, 30 PV, 39 ET, 13 MPN-NOS) recruited at the IRCCS “Giovanni Paolo II” in Bari (2011-2023), the pipeline integrates: (i) Region of Interest (ROI) annotation, (ii) cell segmentation via QuPath/Groovy with extraction of ~9M morphological records, (iii) automated calculation of cellularity, (iv) statistical modeling with a focus on megakaryocytes, (v) linear SVM classification (with 5-fold cross-validation training performed 5 times). The model achieves competitive performance in diagnostic stratification, overcoming inter-observer variability. Validated on a prospective NOS cohort, the workflow supports pathologists in gray scenarios, opening up future deep learning approaches with a view to precision medicine. Translated with DeepL.com (free version)
Le neoplasie mieloproliferative Philadelphia-negative (MPN Ph-) in stadio precoce rappresentano una sfida diagnostica per sovrapposizione morfologica tra policitemia vera (PV), trombocitemia essenziale (ET) e mielofibrosi prefibrotica (pre-PMF). Questa tesi sviluppa una pipeline end-to-end per la classificazione automatica basata esclusivamente su analisi morfometrica di biopsie osteo-midollari digitalizzate (BMB) tramite Whole Slide Imaging (WSI). Su una coorte retrospettiva di 101 BMB (19 pre-PMF, 30 PV, 39 ET, 13 MPN-NOS) reclutata all'IRCCS "Giovanni Paolo II" di Bari (2011-2023), la pipeline integra: (i) annotazione di Region of Interest (ROI), (ii) segmentazione cellulare via QuPath/Groovy con estrazione di ~9M records morfologici, (iii) calcolo automatizzato della cellularità, (iv) modellazione statistica con focus su megacariociti, (v) classificazione SVM lineare ( con addestramento in cross validation con 5 folds effettuata 5 volte). Il modello raggiunge performance competitive nella stratificazione diagnostica, superando la variabilità inter-osservatore. Validato su coorte prospettica NOS, il workflow supporta i patologi in scenari grigi, aprendo a futuri approcci deep learning nell’ottica della medicina di precisione.
CLASSIFICAZIONE AUTOMATICA DELLE NEOPLASIE MIELOPROLIFERATIVE IN STADIO INIZIALE TRAMITE ANALISI MORFOLOGICA DA IMMAGINI ISTOLOGICHE E ALGORITMI DI MACHINE LEARNING
LUPO, ANDREA
2024/2025
Abstract
Early-stage Philadelphia-negative myeloproliferative neoplasms (Ph-MPN) represent a diagnostic challenge due to morphological overlap between polycythemia vera (PV), essential thrombocythemia (ET), and prefibrotic myelofibrosis (pre-PMF). This thesis develops an end-to-end pipeline for automatic classification based exclusively on morphometric analysis of digitized bone marrow biopsies (BMB) using Whole Slide Imaging (WSI). On a retrospective cohort of 101 BMBs (19 pre-PMF, 30 PV, 39 ET, 13 MPN-NOS) recruited at the IRCCS “Giovanni Paolo II” in Bari (2011-2023), the pipeline integrates: (i) Region of Interest (ROI) annotation, (ii) cell segmentation via QuPath/Groovy with extraction of ~9M morphological records, (iii) automated calculation of cellularity, (iv) statistical modeling with a focus on megakaryocytes, (v) linear SVM classification (with 5-fold cross-validation training performed 5 times). The model achieves competitive performance in diagnostic stratification, overcoming inter-observer variability. Validated on a prospective NOS cohort, the workflow supports pathologists in gray scenarios, opening up future deep learning approaches with a view to precision medicine. Translated with DeepL.com (free version)| File | Dimensione | Formato | |
|---|---|---|---|
|
TESI_da_consegnare_pulita_da_consegnare (1).pdf
non disponibili
Dimensione
2.28 MB
Formato
Adobe PDF
|
2.28 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14253/67316