Does Computer-Aided Diagnosis Permit Differentiation of Angiomyolipoma Without Visible Fat From Renal Cell Carcinoma on MDCT?

Objective: The purpose of this study was to evaluate the diagnostic value of computer-aided diagnosis (CADx) in differentiating angiomyolipoma without visible fat from renal cell carcinoma (RCC) on MDCT.

Materials And Methods: The study included 406 patients who had 47 angiomyolipomas without visible fat and 359 RCCs smaller than 4 cm, all of which were diagnosed on the basis of findings from nephrectomy or percutaneous biopsy performed at our institution between 2000 and 2011. MDCT (slice thickness, 2.5 mm for corticomedullary phase image or 5 mm for the other phase images) and clinical findings were blindly reviewed by two radiologists in a single session. At the time the study was performed, radiologist 1 had 8 years of experience, and radiologist 2 had 18 years of experience. On the basis of the MDCT and clinical findings, CADx classified renal tumors as angiomyolipoma and RCC, and each radiologist independently recorded the probability score (0-5) for angiomyolipoma. The accuracy of CADx versus radiologists in diagnosing angiomyolipoma was compared using ROC analysis. Interobserver agreement between the two radiologists was evaluated.

Results: CADx yielded an area under the curve (Az) value of 0.949, which was greater than the Az values yielded by radiologists 1 and 2 (0.872 and 0.782, respectively; p < 0.05). In addition, the Az value for radiologist 1 was greater than that for radiologist 2 (p = 0.01). CADx with a threshold of -1.0085 showed greater sensitivity than radiologist 1 and greater sensitivity, specificity, and accuracy than radiologist 2 (p < 0.05). The interobserver agreement for the differentiation was fair (κ = 0.289).

Conclusion: CAD can improve diagnostic performance in differentiating angiomyolipoma from RCC. The diagnostic performance of radiologists is variable according to the clinical experience and physical and emotional states of the radiologists.