G-M-Y Zhang1, H Sun2, B Shi3, M Xu4, H-D Xue5, Z-Y Jin6. 1. Department of Radiology, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences. Shuaifuyuan No.1, Wangfujing Street, Dongcheng District, Beijing 100730, China. 2. Department of Radiology, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences. Shuaifuyuan No.1, Wangfujing Street, Dongcheng District, Beijing 100730, China. Electronic address: sunhao_robert@126.com. 3. Department of Radiology, Shenzhen Sun Yat-Sen Cardiovascular Hospital, No. 1021 Dongmen Road North, Luohu District, Shenzhen 518001, China. 4. Siemens Healthcare Ltd, Beijing, China. No.7 Zhonghuan Nanlu, Chaoyang District, Beijing 100102, China. 5. Department of Radiology, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences. Shuaifuyuan No.1, Wangfujing Street, Dongcheng District, Beijing 100730, China. Electronic address: bjdanna95@hotmail.com. 6. Department of Radiology, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences. Shuaifuyuan No.1, Wangfujing Street, Dongcheng District, Beijing 100730, China. Electronic address: jinzy@pumch.cn.
Abstract
AIM: To evaluate the accuracy of computed tomography (CT) texture analysis (TA) to differentiate uric acid (UA) stones from non-UA stones on unenhanced CT in patients with urinary calculi with ex vivo Fourier transform infrared spectroscopy (FTIR) as the reference standard. MATERIALS AND METHODS: Fourteen patients with 18 UA stones and 31 patients with 32 non-UA stones were included. All the patients had preoperative CT evaluation and subsequent surgical removal of the stones. CTTA was performed on CT images using commercially available research software. Each texture feature was evaluated using the non-parametric Mann-Whitney test. Receiver operating characteristic (ROC) curves were created and the area under the ROC curve (AUC) was calculated for texture parameters that were significantly different. The features were used to train support vector machine (SVM) classifiers. Diagnostic accuracy was evaluated. RESULTS: Compared to non-UA stones, UA stones had significantly lower mean, standard deviation and mean of positive pixels but higher kurtosis (p<0.001) on both unfiltered and filtered texture scales. There were no significant differences in entropy or skewness between UA and non-UA stones. The average SVM accuracy of texture features for differentiating UA from non-UA stones ranged from 88% to 92% (after 10-fold cross validation). A model incorporating standard deviation, skewness, and kurtosis from unfiltered texture scale images resulted in an AUC of 0.965±00.029 with a sensitivity of 94.4% and specificity of 93.7%. CONCLUSION: CTTA can be used to accurately differentiate UA stones from non-UA stones in vivo using unenhanced CT images.
AIM: To evaluate the accuracy of computed tomography (CT) texture analysis (TA) to differentiate uric acid (UA) stones from non-UA stones on unenhanced CT in patients with urinary calculi with ex vivo Fourier transform infrared spectroscopy (FTIR) as the reference standard. MATERIALS AND METHODS: Fourteen patients with 18 UA stones and 31 patients with 32 non-UA stones were included. All the patients had preoperative CT evaluation and subsequent surgical removal of the stones. CTTA was performed on CT images using commercially available research software. Each texture feature was evaluated using the non-parametric Mann-Whitney test. Receiver operating characteristic (ROC) curves were created and the area under the ROC curve (AUC) was calculated for texture parameters that were significantly different. The features were used to train support vector machine (SVM) classifiers. Diagnostic accuracy was evaluated. RESULTS: Compared to non-UA stones, UA stones had significantly lower mean, standard deviation and mean of positive pixels but higher kurtosis (p<0.001) on both unfiltered and filtered texture scales. There were no significant differences in entropy or skewness between UA and non-UA stones. The average SVM accuracy of texture features for differentiating UA from non-UA stones ranged from 88% to 92% (after 10-fold cross validation). A model incorporating standard deviation, skewness, and kurtosis from unfiltered texture scale images resulted in an AUC of 0.965±00.029 with a sensitivity of 94.4% and specificity of 93.7%. CONCLUSION:CTTA can be used to accurately differentiate UA stones from non-UA stones in vivo using unenhanced CT images.