In recent years, Radiomics has found wide development in image processing, mainly in the oncology field. Radiomics was defined as the high-throughput extraction of quantitative features from medical images such as Positron Emission Tomography (PET), Computed Tomography and Magnetic Resonance (MR) to characterize cancer. Radiomics has shown potential in "precision medicine" whose purpose is to optimize and customize the treatment on the specific patient. This work aims to evaluate some critical issues related to radiomic analysis by using a multimodal anthropomorphic phantom. Phantoms have always been used in medical imaging both in clinical routine to carry out the quality controls to ensure the correct functioning of scanners, and in the research, mainly in the oncology field, with the aim of accurately evaluating, with a known Gold Standard, image processing approaches. Several phantoms have been developed. These phantoms are compatible, in most of the cases, with a single imaging technique. Furthermore, their design present inserts with defined geometric shapes, thus they do not simulate realistic condition of oncological patient. The first aim of this thesis was therefore to develop a method to set up anthropomorphic phantoms simulating the thoracic-abdominal district of a patient with realistic oncological lesions. The phantom was developed in order to compatible with both Positron Emission Tomography (PET) imaging both with Magnetic Resonance Imaging (MR) was considered. A commercial anthropomorphic phantom simulating the thoracic-abdominal district with its organs was used. The 3D printer was used to simulate realistic oncological lesions whit irregular shape obtained from the segmentation of neoplastic lesions of real patients.3D printed shells can be filled with an agar-based gel to simulate signal and contrast of real cancer lesions in PET and MR images. Calibration measurements were made to optimize the setting of the phantom to simulate different pathological conditions that can be seen in cancer patients. Phantom images were acquired in PET and MR were then quantitatively analyzed using a radiomics approach. The oncological lesions were segmented, and quantitative imaging features were extracted from segmented VOIs representing oncological lesions in both PET and MR images. Different imaging features (IFs) were extracted both as local IFs measured as average values within VOI and from the analysis of morphology, histogram and texture of VOIs. Some critical issues of the radiomic analysis pipeline were evaluated, such as the impact of segmentation and the dependence on image reconstruction and pre-processing parameters. The analysis allows to evaluate and develop robust and stable methods for extracting and selecting radiomic features, to be proposed for radiomics analysis in cancer patient studies.
Negli ultimi anni la Radiomica ha trovato ampio sviluppo nell’image processing, principalmente in campo oncologico. L'estrazione e l'analisi di caratteristiche (“features”) provenienti da immagini mediche quali la Tomografia ad Emissione di Positroni (PET), la Tomografia Computerizzata (TC) e la Risonanza Magnetica (MR) hanno reso la radiomica un’interessante metodologia alla base della "medicina di precisione" la cui finalità consiste nel personalizzare il trattamento sulle caratteristiche specifiche del paziente e della sua malattia. Il presente lavoro di tesi mira ad approfondire alcuni aspetti fondamentali che ancora rimangono critici relativi all’analisi radiomica, studiandoli grazie all’uso di un fantoccio antropomorfo multimodale. I fantocci sono sempre stati utilizzati nell’imaging medicale sia nella routine clinica per effettuare i controlli di qualità necessari a garantire il corretto funzionamento delle apparecchiature, sia nel campo della ricerca, principalmente in ambito oncologico, con l’obiettivo di valutare accuratamente, tramite un Gold Standard conosciuto e affidabile, gli approcci di image processing non valutabili in modo quantificabile e obiettivo nel paziente. Diversi fantocci sono stati sviluppati in modo da risultare compatibili con diverse modalità di imaging diagnostico. Nella maggior parte dei casi si tratta di fantocci compatibili con una sola tecnica di imaging e poco realistici poiché presentano inserti dalle forme geometriche definite. Primo obiettivo di questo lavoro di tesi è stato dunque sviluppare una metodica che consentisse di realizzare fantocci antropomorfi, in grado di simulare il distretto toracico-addominale di un paziente con lesioni oncologiche realistiche. Nel valutare come realizzare il fantoccio è stata considerata la possibilità di configurarlo in modo da essere compatibile sia con l’imaging di Tomografia ad Emissione di Positroni (PET); sia con l’imaging di Risonanza Magnetica (MR). E’ stato utilizzato un fantoccio antropomorfo commerciale simulante il distretto toracico-addominale con i suoi organi. Per simulare lesioni oncologiche realistiche, la cui forma irregolare è stata ottenuta dalla segmentazione di lesioni neoplastiche di pazienti reali, è stata utilizzata la stampante 3D. Sono stati quindi realizzati dei gusci riempibili con un gel a base agar in grado di simulare nelle immagini PET e MR, il segnale ed il contrasto proprio di lesioni oncologiche reali. Misure di calibrazione hanno permesso di ottimizzare la preparazione del fantoccio e di valutare la composizione del gel per poter effettuare diverse misure in fantoccio, che potessero simulare diverse situazioni di patologia effettivamente riscontrate nei pazienti oncologici. Le immagini del fantoccio che sono state acquisite in PET e MR sono state poi analizzate quantitativamente usando un approccio di radiomica. Le lesioni oncologiche sono state segmentate e dalle segmentazioni sono state estratte caratteristiche quantitative sia dell’analisi dei valori medi del segnale, sia dall’analisi dell’istogramma sia dall’analisi della tessitura dell’immagine, in particolare per le acquisizioni PET e per le acquisizioni MR pesate T2 e in diffusione. Sono stati valutati alcuni aspetti critici della pipeline di analisi radiomica quali la segmentazione e la dipendenza dai parametri di ricostruzione e di pre-processing delle immagini, al fine di valutare e mettere a punto metodi robusti e stabili per l’estrazione e la selezione delle caratteristiche quantitative, da proporre per le analisi di radiomica in studi con pazienti oncologici.
Sviluppo di un fantoccio multimodale per la valutazione, l’ottimizzazione e la standardizzazione dell’analisi radiomica nell’imaging PET e MR.
D'ARCANGELO, MICOL
2019/2020
Abstract
In recent years, Radiomics has found wide development in image processing, mainly in the oncology field. Radiomics was defined as the high-throughput extraction of quantitative features from medical images such as Positron Emission Tomography (PET), Computed Tomography and Magnetic Resonance (MR) to characterize cancer. Radiomics has shown potential in "precision medicine" whose purpose is to optimize and customize the treatment on the specific patient. This work aims to evaluate some critical issues related to radiomic analysis by using a multimodal anthropomorphic phantom. Phantoms have always been used in medical imaging both in clinical routine to carry out the quality controls to ensure the correct functioning of scanners, and in the research, mainly in the oncology field, with the aim of accurately evaluating, with a known Gold Standard, image processing approaches. Several phantoms have been developed. These phantoms are compatible, in most of the cases, with a single imaging technique. Furthermore, their design present inserts with defined geometric shapes, thus they do not simulate realistic condition of oncological patient. The first aim of this thesis was therefore to develop a method to set up anthropomorphic phantoms simulating the thoracic-abdominal district of a patient with realistic oncological lesions. The phantom was developed in order to compatible with both Positron Emission Tomography (PET) imaging both with Magnetic Resonance Imaging (MR) was considered. A commercial anthropomorphic phantom simulating the thoracic-abdominal district with its organs was used. The 3D printer was used to simulate realistic oncological lesions whit irregular shape obtained from the segmentation of neoplastic lesions of real patients.3D printed shells can be filled with an agar-based gel to simulate signal and contrast of real cancer lesions in PET and MR images. Calibration measurements were made to optimize the setting of the phantom to simulate different pathological conditions that can be seen in cancer patients. Phantom images were acquired in PET and MR were then quantitatively analyzed using a radiomics approach. The oncological lesions were segmented, and quantitative imaging features were extracted from segmented VOIs representing oncological lesions in both PET and MR images. Different imaging features (IFs) were extracted both as local IFs measured as average values within VOI and from the analysis of morphology, histogram and texture of VOIs. Some critical issues of the radiomic analysis pipeline were evaluated, such as the impact of segmentation and the dependence on image reconstruction and pre-processing parameters. The analysis allows to evaluate and develop robust and stable methods for extracting and selecting radiomic features, to be proposed for radiomics analysis in cancer patient studies.È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/11916