Taryn Hodgdon MD, Presenter: Nothing to Disclose

Rebecca Thornhill, Abstract Co-Author: Nothing to Disclose

Matthew Donald Fernand McInnes MD, FRCPC, Abstract Co-Author: Nothing to Disclose

Nicola Schieda MD, Abstract Co-Author: Nothing to Disclose

Leslie Lamb MD, MSc, Abstract Co-Author: Nothing to Disclose

Trevor A. Flood MD, FRCPC, Abstract Co-Author: Nothing to Disclose

To evaluate the accuracy of texture analysis to differentiate mfAML from RCC on non-enhanced computed tomography (NECT), using histopathologic diagnosis of surgically resected renal lesions as the reference standard.

A retrospective case-control study was approved by the institutional review board. Patients with AML and RCC were obtained from the pathology database of surgically resected specimens from January 2002 to August 2013. The study included 16 patients with mfAML and 68 patients with RCC. mfAML was defined by the absence of visible fat on NECT. Preoperative NECTs were reviewed, and texture analysis was performed on 3 axial images of each renal lesion. The most discriminative features were used to generate a support vector machine (SVM) classifier. Accuracy of the SVM was then determined by 10-fold cross validation. The NECT for each patient was also independently reviewed by two blinded radiologists who subjectively graded lesion heterogeneity. The diagnostic performance of textural classifiers was compared with radiologist ratings using McNemar tests.

CT texture features related to lesion homogeneity and entropy were evaluated. There was significantly lower lesion homogeneity and higher lesion entropy in RCC compared to mfAML (p<0.0001 and p = 0.0001 respectively). Each logistic regression model produced by combining various groups of texture features yielded AUC significantly greater than 0.5 (ranging from 0.84-0.87). The model incorporating all five analyzed texture classifiers resulted in an AUC of 0.87, and was able to correctly identify RCC with 79% sensitivity and 81% specificity (with SVM accuracy of 81). Subjective lesion heterogeneity on NECT showed moderate interobserver agreement, with intraclass correlation (ICC) of 0.47 (95% CI 0.27 – 0.62). Optimal heterogeneity rating of greater than 2 was identified as a predictor of RCC for both readers (sensitivity 47%, specificity 88% for reader 1 and sensitivity 56%, specificity 75% for reader 2 ). The combined textural classifiers were significantly more accurate than both radiologists subjective heterogeneity ratings for predicting a diagnosis of RCC.

CT texture analysis may be useful for differentiating mfAML from RCC on NECT.

CT texture analysis features related to lesion homogeneity and entropy may be useful for differentiating mfAML from RCC on NECT.

Hodgdon, T, Thornhill, R, McInnes, M, Schieda, N, Lamb, L, Flood, T, Quantitative CT Texture Analysis: Can It Differentiate between Minimal Fat Renal Angiomyolipoma (mfAML) and Renal Cell Carcinoma on Non-contrast Enhanced Computed Tomography (NECT)?.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14009586.html