Breast-conserving surgery requires the complete removal of the tumour together with a healthy margin of about 2 mm, yet incomplete resections still occur in 5–20% of procedures and frequently lead to re-intervention. This thesis, carried out within the European Spectra-BREAST project, describes the development of a hyperspectral imaging system for the identification of tumour tissue on ex vivo surgical specimens. The system is intended to guide a subsequent confirmatory Raman measurement performed by a robotic arm. It is built around a Fourier-transform hyperspectral camera based on a common-path birefringent interferometer, which acquires the full 400–1700 nm range over the entire field of view in a single measurement, without spatial scanning. Because the spectral content of the data is determined entirely by the illumination, a multichannel LED illuminator combined with a halogen lamp was characterised and optimised to provide a uniform spectrum across the whole band. Surface glare is usually corrected after acquisition; here specular highlights were instead suppressed at the hardware level by a white enclosure providing purely diffuse illumination, which reduces the residual surface reflection to a nearly uniform offset that is removed by normalisation. Thermal measurements showed a specimen temperature increase of about 2 ◦C during acquisition, which is compatible with tissue preservation. The system was validated and installed permanently at the pathology laboratory of IRCCS Maugeri, where it is operated independently by hospital staff. A preliminary analysis on two patients used Standard Normal Variate normalisation, Savitzky–Golay smoothing and Principal Component Analysis. It showed that tumour and adipose tissue are separated by a single principal component, dominated by the lipid and water absorption bands in the near- and short-wave infrared, and that this separation is preserved from one patient to another. Collaborations with the project partners on 3D specimen mapping, depth contrast and histological coregistration are also discussed.
La chirurgia conservativa al seno richiede la rimozione completa del tumore insieme a un margine di tessuto sano di circa 2 mm. Tuttavia, le resezioni incomplete si verif- icano ancora nel 5–20% degli interventi e spesso rendono necessaria una seconda oper- azione. Questa tesi, svolta nell’ambito del progetto europeo Spectra-BREAST, descrive lo sviluppo di un sistema di imaging iperspettrale per l’identificazione del tessuto tumorale su campioni chirurgici ex vivo. Il sistema ha lo scopo di guidare una successiva misura Raman di conferma, eseguita da un braccio robotico. Si basa su una camera iperspettrale a trasformata di Fourier con interferometro birifrangente a cammino comune, che ac- quisisce l’intero intervallo 400–1700 nm su tutto il campo di vista in un’unica misura, senza scansione spaziale. Poiché il contenuto spettrale dei dati dipende interamente dall’illuminazione, è stato caratterizzato e ottimizzato un illuminatore LED multicanale, combinato con una lampada alogena, per ottenere uno spettro uniforme su tutta la banda. Il glare superficiale viene solitamente corretto dopo l’acquisizione; qui i riflessi speculari sono stati invece soppressi a livello hardware tramite una struttura bianca che garantisce un’illuminazione puramente diffusa, riducendo la riflessione superfiale residua a un off- set pressoché uniforme, rimosso in fase di normalizzazione. Le misure termiche hanno mostrato un aumento di temperatura del campione di circa 2 ◦C durante l’acquisizione, compatibile con la sua conservazione. Il sistema è stato validato e installato in modo permanente presso il laboratorio di anatomia patologica dell’IRCCS Maugeri, dove è uti- lizzato autonomamente dal personale ospedaliero. Un’analisi preliminare su due pazienti ha impiegato la normalizzazione Standard Normal Variate, il filtro di Savitzky–Golay e l’Analisi delle Componenti Principali. Ha mostrato che tessuto tumorale e adiposo sono separati da una singola componente principale, dominata dalle bande di assorbimento di lipidi e acqua nel vicino infrarosso e nell’infrarosso a onda corta, e che tale separazione si mantiene tra pazienti diversi. Sono infine discusse le collaborazioni con i partner su mappatura 3D dei campioni, contrasto in profondità e coregistrazione istologica.
Spectra-breast: hyperspectral imaging for tumour tissue identification in ex-vivo specimens
PERONI, ANDREA
2025/2026
Abstract
Breast-conserving surgery requires the complete removal of the tumour together with a healthy margin of about 2 mm, yet incomplete resections still occur in 5–20% of procedures and frequently lead to re-intervention. This thesis, carried out within the European Spectra-BREAST project, describes the development of a hyperspectral imaging system for the identification of tumour tissue on ex vivo surgical specimens. The system is intended to guide a subsequent confirmatory Raman measurement performed by a robotic arm. It is built around a Fourier-transform hyperspectral camera based on a common-path birefringent interferometer, which acquires the full 400–1700 nm range over the entire field of view in a single measurement, without spatial scanning. Because the spectral content of the data is determined entirely by the illumination, a multichannel LED illuminator combined with a halogen lamp was characterised and optimised to provide a uniform spectrum across the whole band. Surface glare is usually corrected after acquisition; here specular highlights were instead suppressed at the hardware level by a white enclosure providing purely diffuse illumination, which reduces the residual surface reflection to a nearly uniform offset that is removed by normalisation. Thermal measurements showed a specimen temperature increase of about 2 ◦C during acquisition, which is compatible with tissue preservation. The system was validated and installed permanently at the pathology laboratory of IRCCS Maugeri, where it is operated independently by hospital staff. A preliminary analysis on two patients used Standard Normal Variate normalisation, Savitzky–Golay smoothing and Principal Component Analysis. It showed that tumour and adipose tissue are separated by a single principal component, dominated by the lipid and water absorption bands in the near- and short-wave infrared, and that this separation is preserved from one patient to another. Collaborations with the project partners on 3D specimen mapping, depth contrast and histological coregistration are also discussed.| File | Dimensione | Formato | |
|---|---|---|---|
|
Executive_Summary__Master_Thesis.pdf
non accessibile
Dimensione
9.17 MB
Formato
Adobe PDF
|
9.17 MB | Adobe PDF | Visualizza/Apri |
|
SpectraBREAST Hyperspectral Imaging for Tumour Tissue Identification in Ex Vivo Surgical Specimens.pdf
non accessibile
Dimensione
80.4 MB
Formato
Adobe PDF
|
80.4 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/10589/267139