Quantitative methods for the assessment of intestitial lung disease in MDCT

Interstitial lung diseases are a heterogeneous group of disorders that vary widely in etiology, clinic-radiologic presentation, histopathologic features, and clinical course. MDCT is the modality of choice for determining the extent of diffuse interstitial lung disease and predicting the clinical...

Πλήρης περιγραφή

Λεπτομέρειες βιβλιογραφικής εγγραφής
Κύριος συγγραφέας: Παπαπαναγιώτου, Νικόλαος
Άλλοι συγγραφείς: Κωσταρίδου, Ελένη
Μορφή: Thesis
Γλώσσα:English
Έκδοση: 2014
Θέματα:
Διαθέσιμο Online:http://hdl.handle.net/10889/7784
Περιγραφή
Περίληψη:Interstitial lung diseases are a heterogeneous group of disorders that vary widely in etiology, clinic-radiologic presentation, histopathologic features, and clinical course. MDCT is the modality of choice for determining the extent of diffuse interstitial lung disease and predicting the clinical outcomes as the scoring of fibrosis correlates well with the mortality rate. Different visual scoring systems for evaluating ILDs’ extent on HRCT have been developed over the past 20 years. Several visual scoring methods have been used to characterize and quantify the disease, correlate with common clinical parameters, prognosticate patients, assess disease progression and evaluate response to treatment. Up to date, visual scoring remains the method of choice for assessing disease extent in clinical practice. However, these methods show variable reproducibility in literature and therefore, a more accurate classification system is necessary for objective and reproducible assessment of disease extent. This has lead to considerable research efforts in advanced computer-based ILD extent quantification systems in the last 10 years. In this Thesis we compare four different available methods for the assessment of interstitial lung disease, for total, ground glass and reticular extent. A radiologist in training evaluated disease extent using a semi-quantitative visual scoring method (a), a visual pixel-based method (b) and semi-automated histogram thresholding technique (c). An automated CAD algorithm (d) was also utilized. All methods were applied to the same data sample of patients with collagen vascular diseases and lung involvement. The sample performance is reported on axial slice basis in terms of mean, standard deviation and range. Furthermore, methods have been compared pairwise by means of Bland-Altman analysis, utilized in order to assess by inspection the degree of agreement for varying disease extent. Additionally, the Intraclass Correlation Coefficient index has been calculated for all pairs compared. Statistical analysis showed almost perfect agreement between our visual pixel based method and the automated system concerning total and reticular disease extent, while the CAD algorithm and thresholding technique have demonstrated substantial agreement. None of the pairwise comparisons exhibited a high degree of agreement concerning ground glass extent estimation. Inter-observer comparison manifested significantly higher degree of agreement for the visual pixel based technique as compared to semi-quantitative visual scoring method. CAD algorithms provide a fast and reproducible disease extent and in our study present a high agreement with visual pixel based method, which is accounted for the more precise, albeit time wasting method. Resultantly, these automated systems could replace semi-quantitative visual scoring methods, not radiologists, in terms of accuracy, reproducibility and more precise clinical decision.