Breast cancer is one of the most frequently diagnosed diseases in women, alternating in recent years among the top four cancer-related causes of death. Its rising incidence highlights the need for diagnostic techniques that are both accurate and non-invasive. Traditional imaging modalities, such as mammography, ultrasound, MRI and PET, are widely used but have limitations in terms of invasiveness, costs and lack of quantitative information. Within this framework, the SOLUS project proposes a multimodal diagnostic strategy that combines time-domain diffuse optical imaging with ultrasound methods, including B-mode, Shear-Wave Elastography, and Color Doppler, with the aim of improving lesion diagnosis and reducing unnecessary biopsies. This thesis, part of the ongoing SOLUS project, investigates the performance of the novel system through measurements on phantoms and the analysis of both these and in vivo data from 40 patients (with benign and malignant lesions) collected in a different project phase. As part of this work, structured databases were also developed to organize the information prior to systematic in vivo analysis and prepare for open data publication. Two data analysis pipelines were explored for in vivo cases. The analytical approach involved the extraction of optical parameters, including absorption coefficients and chromophore concentrations (oxy- deoxy-hemoglobin, water, lipids, collagen), through Diffuse Optical Tomography (DOT), followed by descriptive and inferential statistical analysis, as well as machine learning classification. The direct approach relied mainly on raw distributions of photon time-of-flight (DTOF) data, the typical output of a time-resolved diffuse optical system, adopting a more customized machine learning framework tailored to the characteristics of the signals. Statistical results are meaningful, and classification performance is promising, especially when optical and ultrasound descriptive information are integrated. This work provides a contribution to the evaluation of the SOLUS system and supports ongoing developments toward more accurate, quantitative, and accessible breast cancer diagnostics.
Il tumore al seno è una delle patologie più frequentemente diagnosticate nelle donne, alternandosi negli ultimi anni tra le prime quattro cause di morte per cancro. L’aumento della sua incidenza evidenzia la necessità di tecniche di diagnosi accurate e non invasive. I metodi di imaging tradizionale, come mammografia, ecografia, MRI e PET, sono ampiamente utilizzati, sebbene con limitazioni in termini di invasività, costi e assenza di quantitatività. In tale contesto, il progetto SOLUS propone una strategia diagnostica multimodale che combina l’imaging ottico in dominio del tempo con metodiche ecografiche, tra cui B-mode, Shear-Wave Elastography e Color Doppler, allo scopo di migliorare la diagnosi delle lesioni e ridurre le biopsie non necessarie. Questa tesi, svolta all’interno di tale progetto, valuta le prestazioni del sistema innovativo tramite misure su phantom e l’analisi sia di questi dati che di dati in vivo, relativi a 40 pazienti (con lesioni benigne e maligne) raccolti in una fase precedente. Grazie a questo lavoro sono stati inoltre sviluppati database strutturati, utili a organizzare le informazioni per l’analisi sistematica in vivo e in vista della loro condivisione pubblica. Sono stati esplorati due approcci di analisi per i casi in vivo. L’approccio analitico ha previsto l’estrazione dei parametri ottici, come i coefficienti di assorbimento e le concentrazioni dei cromofori (ossi-/ desossi- emoglobina, acqua, lipidi, collagene) tramite tomografia ottica diffusa (DOT), seguita da analisi statistiche descrittive, inferenziali e da classificazione con machine learning. L’approccio diretto si è basato specialmente sulle distribuzioni grezze dei tempi di volo dei fotoni (DTOF), output di un sistema ottico diffuso risolto in tempo, adottando un’architettura di machine learning su misura per le caratteristiche dei segnali. I risultati statistici sono significativi e le prestazioni di classificazione promettenti, soprattutto integrando le informazioni ottiche con quelle ecografiche. Questo lavoro fornisce un contributo alla valutazione del sistema SOLUS e sostiene gli sviluppi di una diagnostica del tumore al seno più accurata, quantitativa e accessibile.
Breast cancer classification through diffuse optics and ultrasound: solus system testing and data analysis
Ayvazian, Maria Vittoria
2024/2025
Abstract
Breast cancer is one of the most frequently diagnosed diseases in women, alternating in recent years among the top four cancer-related causes of death. Its rising incidence highlights the need for diagnostic techniques that are both accurate and non-invasive. Traditional imaging modalities, such as mammography, ultrasound, MRI and PET, are widely used but have limitations in terms of invasiveness, costs and lack of quantitative information. Within this framework, the SOLUS project proposes a multimodal diagnostic strategy that combines time-domain diffuse optical imaging with ultrasound methods, including B-mode, Shear-Wave Elastography, and Color Doppler, with the aim of improving lesion diagnosis and reducing unnecessary biopsies. This thesis, part of the ongoing SOLUS project, investigates the performance of the novel system through measurements on phantoms and the analysis of both these and in vivo data from 40 patients (with benign and malignant lesions) collected in a different project phase. As part of this work, structured databases were also developed to organize the information prior to systematic in vivo analysis and prepare for open data publication. Two data analysis pipelines were explored for in vivo cases. The analytical approach involved the extraction of optical parameters, including absorption coefficients and chromophore concentrations (oxy- deoxy-hemoglobin, water, lipids, collagen), through Diffuse Optical Tomography (DOT), followed by descriptive and inferential statistical analysis, as well as machine learning classification. The direct approach relied mainly on raw distributions of photon time-of-flight (DTOF) data, the typical output of a time-resolved diffuse optical system, adopting a more customized machine learning framework tailored to the characteristics of the signals. Statistical results are meaningful, and classification performance is promising, especially when optical and ultrasound descriptive information are integrated. This work provides a contribution to the evaluation of the SOLUS system and supports ongoing developments toward more accurate, quantitative, and accessible breast cancer diagnostics.File | Dimensione | Formato | |
---|---|---|---|
2025_07_Ayvazian_Tesi_01.pdf
non accessibile
Descrizione: Tesi
Dimensione
18.21 MB
Formato
Adobe PDF
|
18.21 MB | Adobe PDF | Visualizza/Apri |
2025_07_Ayvazian_Executive_Summary_02.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
1.05 MB
Formato
Adobe PDF
|
1.05 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/240529