Lung tumour is the leading cause of cancer-related mortality worldwide. Pulmonary lobectomy, which consists in the surgical excision of the injured lobe, is performed in patients with early stage cancer that are suitable for receiving operation. The preoperative pulmonary function is assessed by spirometry, measuring the forced expiratory volume in 1 second (FEV1), and the predicted postoperative FEV1 (ppoFEV1) is traditionally estimated by the proportion of lung segments that will be resected to those that will be preserved. Since almost every patient with lung cancer undergoes computed tomography (CT) scanning for diagnosis and treatment planning, extracting quantitative parameters from CT images could improve the post-surgical outcome prediction without the need for additional examinations. The purpose of the present thesis is to evaluate if texture analysis applied to CT images from patients with lung cancer can improve the prediction of postoperative pulmonary function. To perform a quantitative study of high-resolution computed tomography (HRCT) images, a fully automatic algorithm for 3D lung segmentation is developed, based on intrinsic image decomposition, the wavelet transform, and final contour correction by convex hull. Comparing to reference segmentations, Dice similarity coefficient (DSC) and Jaccard similarity index (JSI) achieve an average value of 98% and 96%, respectively, on 19 patients with both increased and decreased attenuation patterns. Histogram-derived texture parameters and pulmonary volumes are then computed from the lung parenchyma segmented with the proposed method, from the affected lobe, and from the remaining lobes of 12 patients. The relationship between the derived measures and preoperative FEV1 and postoperative FEV1 is assessed by statistical analysis. Significant correlations are found between volume of the overall lung parenchyma and preFEV1 (r = 0.80, p = 0.0004) and between remaining lung volume (r = 0.65, p = 0.02), functional remaining volume (r = 0.80, p = 0.0004), and weight of the remaining lobes (r = 0.80, p = 0.0007) and postFEV1.
Il tumore al polmone è la principale causa di mortalità legata al cancro a livello mondiale. La lobectomia polmonare, che consiste nella rimozione chirurgica del lobo interessato, è impiegata nei soggetti con cancro allo stadio iniziale che risultano essere adatti a ricevere l’operazione. La funzione polmonare preoperatoria è valutata tramite spirometria, misurando il volume espiratorio forzato in 1 secondo (FEV1), e il FEV1 postoperatorio viene stimato dal rapporto tra i segmenti polmonari che saranno rimossi e quelli che verranno preservati nell’intervento (ppoFEV1). Poiché quasi tutti i pazienti con cancro ai polmoni vengono sottoposti a tomografia computerizzata (CT) per diagnosi e pianificazione del trattamento, estrarre parametri quantitativi dalle immagini CT potrebbe migliorare la previsione dell’esito post-operatorio senza la necessità di ulteriori esami. Lo scopo del presente lavoro di tesi è valutare l’applicabilità dell’analisi di texture a immagini CT da pazienti con cancro ai polmoni, al fine di migliorare la previsione della funzione polmonare postoperatoria. Per condurre uno studio quantitativo sulle immagini CT ad alta risoluzione (HRCT), un algoritmo completamente automatico per la segmentazione 3D del polmone è sviluppato. L’algoritmo è basato sulla decomposizione intrinseca di immagini, sulla trasformata wavelet, e su una correzione del contorno finale utilizzando l’inviluppo convesso. Confrontando con segmentazioni di riferimento, il coefficiente di similarità di Dice (DSC) e l’indice di similarità di Jaccard (JSI) si attestano rispettivamente ad una media di 98% e 96% su 19 pazienti sia con aree di maggiore attenuazione sia con aree di minore attenuazione. Parametri di texture derivanti dall’istogramma e volumi polmonari sono successivamente calcolati a partire dal parenchima polmonare segmentato con il metodo proposto, dal lobo comprendente il tumore, e dai lobi rimanenti di 12 soggetti. La relazione tra le misure ottenute e il FEV1 preoperatorio e il FEV1 postoperatorio viene valutata attraverso un’analisi statistica. Si rilevano correlazioni significative tra il volume dell’intero parenchima polmonare e il preFEV1 (r = 0.80, p = 0.0004) e tra il volume polmonare rimanente (r = 0.65, p = 0.02), il volume funzionale rimanente (r = 0.80, p = 0.0004), il peso dei lobi rimanenti (r = 0.80, p = 0.0007) e il postFEV1.
Automatic lung segmentation for quantitative CT analysis in patients undergoing pulmonary lobectomy
Molani, Alessandro
2019/2020
Abstract
Lung tumour is the leading cause of cancer-related mortality worldwide. Pulmonary lobectomy, which consists in the surgical excision of the injured lobe, is performed in patients with early stage cancer that are suitable for receiving operation. The preoperative pulmonary function is assessed by spirometry, measuring the forced expiratory volume in 1 second (FEV1), and the predicted postoperative FEV1 (ppoFEV1) is traditionally estimated by the proportion of lung segments that will be resected to those that will be preserved. Since almost every patient with lung cancer undergoes computed tomography (CT) scanning for diagnosis and treatment planning, extracting quantitative parameters from CT images could improve the post-surgical outcome prediction without the need for additional examinations. The purpose of the present thesis is to evaluate if texture analysis applied to CT images from patients with lung cancer can improve the prediction of postoperative pulmonary function. To perform a quantitative study of high-resolution computed tomography (HRCT) images, a fully automatic algorithm for 3D lung segmentation is developed, based on intrinsic image decomposition, the wavelet transform, and final contour correction by convex hull. Comparing to reference segmentations, Dice similarity coefficient (DSC) and Jaccard similarity index (JSI) achieve an average value of 98% and 96%, respectively, on 19 patients with both increased and decreased attenuation patterns. Histogram-derived texture parameters and pulmonary volumes are then computed from the lung parenchyma segmented with the proposed method, from the affected lobe, and from the remaining lobes of 12 patients. The relationship between the derived measures and preoperative FEV1 and postoperative FEV1 is assessed by statistical analysis. Significant correlations are found between volume of the overall lung parenchyma and preFEV1 (r = 0.80, p = 0.0004) and between remaining lung volume (r = 0.65, p = 0.02), functional remaining volume (r = 0.80, p = 0.0004), and weight of the remaining lobes (r = 0.80, p = 0.0007) and postFEV1.File | Dimensione | Formato | |
---|---|---|---|
Master_Thesis_Alessandro_Molani.pdf
solo utenti autorizzati dal 01/04/2022
Dimensione
8.05 MB
Formato
Adobe PDF
|
8.05 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/10589/174932