Premium
Computer‐aided diagnosis of sonographic liver cirrhosis: A spleen‐reference approach
Author(s) -
Yang PeiMing,
Chen ChungMing,
Lu TungWu,
Yen ChihPin
Publication year - 2008
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2870217
Subject(s) - cad , artificial intelligence , ultrasound , robustness (evolution) , receiver operating characteristic , computer aided diagnosis , medical imaging , logistic regression , computer science , pattern recognition (psychology) , medicine , algorithm , cirrhosis , mathematics , radiology , machine learning , biochemistry , chemistry , engineering drawing , gene , gastroenterology , engineering
This article aimed to develop a computer‐aided diagnosis (CAD) algorithm for differential diagnosis of cirrhotic and noncirrhotic livers, allowing free adjustment of system settings to obtain the best scanned images. A spleen‐referenced approached was proposed. 505 pairs of images acquired from 322 subjects using three ultrasound imaging systems were evaluated. Each pair consisted of liver and spleen sonograms of the same subject. Medical doctors were free to adjust the system parameters to attain the best scanned images. Of these 322 subjects, 155 subjects were cirrhotic, and 167 subjects noncirrhotic. Four assessments were carried out to evaluate the effectiveness and robustness of the proposed algorithm on the variations of regions‐of‐interest, ultrasound imaging systems, and scanned images. Four feature pairs were proposed. A logistic regression function was used as the classifier. All performances were evaluated by the leave‐one‐out cross‐validation method. The proposed algorithm was compared with two previous CAD algorithms and all performance comparisons were based on paired‐sample t tests. The meanA zvalues of the receiver operating characteristic curves and classification accuracies ranged as high as from 0.949 to 0.996 and from 91.58 % to 97.30 % , respectively, for the four assessments. At the 5 % family‐wise significance level, the proposed algorithm was shown to be superior to the two previous CAD algorithms and their modified versions in terms of theA zvalues.