Low contrast quality assesment in CT: Human and model observer comparison

• Published in: Physica medica Vol. 32; p. 210
• Main Authors: Biondi, M, LaPenna, A, Vanzi, E, Mazzoni, L.N, De Otto, G, Belmonte, G.M, Foderà, E, Martini, R, Imbriaco, G, Carbone, F.S, Guasti, A, Volterrani, L, Buonamici, F. Banci
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2016
• Subjects: Radiology
• ISSN: 1120-1797; 1724-191X
• DOI: 10.1016/j.ejmp.2016.07.710

Introduction Actually there is not a widely accepted method for low contrast resolution evaluation in CT scans. Purpose To test the performance of a novel ImageJ macro devoted to Catphan CTP515 low contrast automatic analysis by comparing the results with human observers. Materials and methods Software and human proportion correct curves (PCC), function of the low contrast object diameter, and corresponding Minimum Detectable Size (MDS) were determined and compared. Software PCC was obtained using the CTP515 1% contrast group of Catphan 600 images, acquired with Discovery 750 (GE healthcare) 64 slices CT scanner varying noise index (NI = 8, 12) and analysed with a non-prewhitening matched filter with an eye filter model observer. The human PCC was calculate with a known exactly signal and background experiment based on 740 canvas consisting of a template object to be recognized and two images (“background” and “signal + background”) among which the one containing the object had to be chosen. The experiment was repeated with 4 human observers. Inter-observer and human/model PC agreements were evaluated respectively with ICC test and Pearson correlation coefficient with Student t -test. Results MDS for software and human were respectively 2.77 mm and 4.42 mm (12 NI), 1.46 mm and 2.96 mm (8 NI). ICCs were 0.89 (8NI) and 0.84 (12NI); correlation coefficient was 0.89 ( p = 0.02). Conclusion The high correlation between human and model showed that the macro represents a valid tool for low contrast resolution analysis. It could be useful in CT protocol optimization also thanks to its easy use. Disclosure Authors disclose any relationship that may bias their presentation.