Background: Radiomics is a quantitative approach through which medical images are converted into mathematical quantities, namely radiomic features, that have a potential biomedical meaning. Our work aims to preliminarily investigate the multi-messenger power of radiomic features extracted from CT and MRI lung cancer images, and the agreement between two different open-source radiomics software. Methods: CT and MR images from 35 histologically confirmed NSCLC patients were analysed. The images were manually segmented and pre-processed using three different methods, acting on voxels resampling and gray-level discretization. 66 features were extracted, all of which are IBSI-compliant and common to two radiomics platforms, PyRadiomics and LIFEx. Statistical analysis was performed by comparing (i) MRI and CT results (same software), and (ii) Pyradiomics and LIFEx (at fixed technique) in order to investigate the correlation among features with the same mathematical definition. Results: By comparing the same features extracted from LIFEx and PyRadiomics software, statistically significant correlations were found for 94% of CT-derived features and 94% of MR-derived ones, suggesting a high degree of effectiveness of the IBSI standardization. When considering the correlation between features extracted from CT and MR images with the same software, significant correspondences were found for only 15% of LIFEx features and 12% of PyRadiomics features. Conclusion: Our results suggest that: (i) for each imaging technique, IBSI-compliant LIFEx and Pyradiomics software allow to extract more than 90% of features containing the same information; (ii) CT and MR images contain different, non-redundant information, as is expected from the different physical principles governing each modality, which could be used in a complementary way for a multi-modal approach.
Obiettivo: La radiomica è un approccio quantitativo attraverso il quale le immagini mediche vengono convertite in dati matematici, possibilmente con un significato biomedico, denominate features radiomiche. Il nostro lavoro mira a studiare in via preliminare il potere multi-messaggero delle caratteristiche radiomiche estratte da immagini di tomografía computerizzata (TC) e di risonanza magnetica (RM) di pazienti affetti da cancro al polmone e l'accordo tra due diversi software di radiomica open-source. Metodi: sono state analizzate le immagini di 35 pazienti, considerando per ciascuno sia le acquisizioni TC che quelle RM. Le immagini sono state segmentate manualmente e poi pre-processate utilizzando tre metodi diversi, che agiscono sul ricampionamento dei voxel e sulla discretizzazione dei livelli di grigio. Sono state estratte 66 features IBSI-compliant e comuni alle due piattaforme di radiomica, PyRadiomics e LIFEx. La correlazione tra le caratteristiche con la stessa definizione matematica è stata studiata statisticamente confrontando (i) sia i risultati di RM e TC con lo stesso software che quelli di (ii) Pyradiomics e LIFEx su tecnica fissa. Risultati: Confrontando le stesse features estratte dai software LIFEx e PyRadiomics, sono state trovate correlazioni statisticamente significative per il 94% delle features estratte da immagini TC e per il 94% di quelle estratte da immagini RM, suggerendo un elevato grado di efficacia della standardizzazione IBSI. Se si considera la correlazione tra le caratteristiche estratte da immagini TC e RM con lo stesso software, sono state trovate corrispondenze significative solo per il 15% delle caratteristiche LIFEx e per il 12% delle caratteristiche PyRadiomics. Conclusioni: I nostri risultati suggeriscono che: (i) per ogni tecnica di imaging, i software IBSI-compliant LIFEx e Pyradiomics consentono di estrarre più del 90% di features contenenti le stesse informazioni; (ii) le immagini TC e RM contengono informazioni diverse e non ridondanti, come previsto dai diversi principi fisici, e che potrebbero essere utilizzate in modo complementare per un approccio multimodale al cancro al polmone.
MULTI-MESSENGER RADIOMICS OF LUNG CANCER: PRELIMINARY RESULTS FOR RADIOMIC FEATURES STABILITY AMONG DIFFERENT IMAGING MODALITIES AND EXTRACTION SOFTWARE
TIGLIO, CELESTE ALDANA
2022/2023
Abstract
Background: Radiomics is a quantitative approach through which medical images are converted into mathematical quantities, namely radiomic features, that have a potential biomedical meaning. Our work aims to preliminarily investigate the multi-messenger power of radiomic features extracted from CT and MRI lung cancer images, and the agreement between two different open-source radiomics software. Methods: CT and MR images from 35 histologically confirmed NSCLC patients were analysed. The images were manually segmented and pre-processed using three different methods, acting on voxels resampling and gray-level discretization. 66 features were extracted, all of which are IBSI-compliant and common to two radiomics platforms, PyRadiomics and LIFEx. Statistical analysis was performed by comparing (i) MRI and CT results (same software), and (ii) Pyradiomics and LIFEx (at fixed technique) in order to investigate the correlation among features with the same mathematical definition. Results: By comparing the same features extracted from LIFEx and PyRadiomics software, statistically significant correlations were found for 94% of CT-derived features and 94% of MR-derived ones, suggesting a high degree of effectiveness of the IBSI standardization. When considering the correlation between features extracted from CT and MR images with the same software, significant correspondences were found for only 15% of LIFEx features and 12% of PyRadiomics features. Conclusion: Our results suggest that: (i) for each imaging technique, IBSI-compliant LIFEx and Pyradiomics software allow to extract more than 90% of features containing the same information; (ii) CT and MR images contain different, non-redundant information, as is expected from the different physical principles governing each modality, which could be used in a complementary way for a multi-modal approach.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/20.500.14239/16475