Diffuse Optical Tomography (DOT) is an emerging imaging technique experiencing growth and attracting attention from the worldwide research community. This technique promises to achieve quantitative images of the tissue compositions to diagnose lesions in the breast. Despite its potential, DOT development and clinical translation are impaired by the difficulty in providing accurate images due to the diffusive nature of light propagation in tissue, which leads to a severely ill-posed reconstruction problem. Analysis methods for DOT thus play a fundamental role in achieving spendable results and are the focus of our efforts in this work. Hence, we here investigate novel methods and approaches to the analysis of data from SOLUS, a multi-modal multi-wavelength time-domain instrument for ultrasound guided DOT. We achieve a first characterization of the system by the application of the BIP and MEDPHOT protocols to ensure the quality of data to be used thereafter. Then we apply the reconstruction algorithm to obtain images of the absorption and reduced scattering coefficients in heterogeneous meat samples. An analysis of the absorption spectra obtained suggests that the algorithm suffers from a dominating influence of the background, which leads to an inaccurate characterization of the lesion. In this context, we test the application of a pre-processing fit with the homogeneous diffusion model for the reconstruction of the inclusion and background spectra. The results obtained suggest that the method doesn’t alleviate the inaccuracy of the model but can work to denoise data. Finally, we try a different approach and implement Machine Learning algorithms for tissue classification starting from the photon time-of-flight distributions. In this last attempt, to obtain an accurate lesion characterization, we use the preliminary fit of above as a curve regularization technique. The results show good lesion characterization and suggest further research in this direction.
La tomografia ottica diffusa è una tecnica emergente e in piena crescita che si sta affermando nel panorama di ricerca internazionale. Questa tecnica ambisce a ricostruire spazialmente la composizione del tessuto mammario per diagnosticare tumori. Nonostante il grande potenziale, essa soffre di bassa accuratezza nella ricostruzione delle immagini a causa della natura diffusiva dei fotoni all’interno del tessuto, condizione che si traduce in un problema di ricostruzione mal posto. Pertanto, i metodi di analisi in questa tecnologia sono quantomai importanti e rappresentano il focus principale di questo lavoro. Difatti, il nostro obiettivo è l’applicazione di nuove tecniche di analisi sui dati ottenuti con SOLUS, un sistema bimodale di tomografia ottica diffusa guidata da ultrasuoni operante nel dominio del tempo e a diverse lunghezze d’onda. Nel corso di questo lavoro otteniamo una caratterizzazione preliminare dello strumento tramite i protocolli BIP e MEDPHOT, per valutare la qualità dei dati. Successivamente applichiamo l’algoritmo di ricostruzione per ottenere la distribuzione dei coefficienti di assorbimento e scattering in phantom di carne. Un’analisi degli spettri di assorbimento suggerisce che l’algoritmo dia un’importanza dominante alle proprietà ottiche del background, limitando l’accuratezza nella caratterizzazione dell’inclusione. In questo contesto testiamo l'applicazione di un fit preliminare con il modello omogeneo diffusivo per la ricostruzione degli spettri di background e inclusione. I risultati ottenuti suggeriscono che il metodo non è efficace nell’alleviare l’inaccuratezza del modello, ma che potrebbe essere utile per la rimozione del rumore. Infine adottiamo un approccio diverso utilizzando algoritmi di machine learning per la classificazione dei tessuti a partire dalla distribuzione dei tempi d’arrivo dei fotoni. In questo ultimo tentativo di ottenere una caratterizzazione della lesione accurata utilizziamo il fit preliminare per rimuovere il rumore in ingresso. I risultati ottenuti mostrano una classificazione abbastanza accurata e suggeriscono l’utilità di ulteriori studi in questa direzione.
Analysis methods for performance enhancement of SOLUS, a system for diffuse optical tomography of breast lesions
Carella, Alessandro, Pio
2023/2024
Abstract
Diffuse Optical Tomography (DOT) is an emerging imaging technique experiencing growth and attracting attention from the worldwide research community. This technique promises to achieve quantitative images of the tissue compositions to diagnose lesions in the breast. Despite its potential, DOT development and clinical translation are impaired by the difficulty in providing accurate images due to the diffusive nature of light propagation in tissue, which leads to a severely ill-posed reconstruction problem. Analysis methods for DOT thus play a fundamental role in achieving spendable results and are the focus of our efforts in this work. Hence, we here investigate novel methods and approaches to the analysis of data from SOLUS, a multi-modal multi-wavelength time-domain instrument for ultrasound guided DOT. We achieve a first characterization of the system by the application of the BIP and MEDPHOT protocols to ensure the quality of data to be used thereafter. Then we apply the reconstruction algorithm to obtain images of the absorption and reduced scattering coefficients in heterogeneous meat samples. An analysis of the absorption spectra obtained suggests that the algorithm suffers from a dominating influence of the background, which leads to an inaccurate characterization of the lesion. In this context, we test the application of a pre-processing fit with the homogeneous diffusion model for the reconstruction of the inclusion and background spectra. The results obtained suggest that the method doesn’t alleviate the inaccuracy of the model but can work to denoise data. Finally, we try a different approach and implement Machine Learning algorithms for tissue classification starting from the photon time-of-flight distributions. In this last attempt, to obtain an accurate lesion characterization, we use the preliminary fit of above as a curve regularization technique. The results show good lesion characterization and suggest further research in this direction.File | Dimensione | Formato | |
---|---|---|---|
2024_07_Carella_Tesi.pdf
non accessibile
Dimensione
3.5 MB
Formato
Adobe PDF
|
3.5 MB | Adobe PDF | Visualizza/Apri |
2024_07_Carella_Executive Summary.pdf
non accessibile
Dimensione
800.65 kB
Formato
Adobe PDF
|
800.65 kB | 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/10589/223456