Diagnosis of Sarcomatoid Renal Cell Carcinoma With CT: Evaluation by Qualitative Imaging Features and Texture Analysis

• Published in: American journal of roentgenology (1976) Vol. 204; no. 5; pp. 1013 - 1023
• Main Authors: Schieda, Nicola, Thornhill, Rebecca E, Al-Subhi, Maali, McInnes, Matthew D F, Shabana, Wael M, van der Pol, Christian B, Flood, Trevor A
• Format: Journal Article
• Language: English
• Published: United States 01-05-2015
• Subjects: Aged; Carcinoma, Renal Cell - diagnostic imaging; Case-Control Studies; Contrast Media; Female; Humans; Kidney Neoplasms - diagnostic imaging; Male; Middle Aged; Neoplasm Invasiveness; Radiographic Image Interpretation, Computer-Assisted; Retrospective Studies; Tomography, X-Ray Computed - methods; Treatment Outcome; CT; renal cell carcinoma; sarcomatoid; clear cell; texture analysis
• ISSN: 0361-803X; 1546-3141
• DOI: 10.2214/AJR.14.13279

The objective of our study was to determine whether CT findings, including texture analysis, can differentiate sarcomatoid renal cell carcinoma (RCC) from clear cell RCC.
A retrospective case-control study was performed of consecutive patients with a histologic diagnosis of sarcomatoid RCC (n = 20) and clear cell RCC (n = 25) who underwent preoperative CT over a 3-year period. The CT images were independently reviewed by two blinded abdominal radiologists; they evaluated the following: tumor heterogeneity, tumor margin, calcification, intratumoral neovascularity, peritumoral neovascularity, renal sinus invasion, renal vein invasion, and adjacent organ invasion. Interobserver agreement was assessed using the Cohen kappa coefficient, and results were compared between groups using an independent Student t test and the chi-square test with a Bonferroni correction. For texture analysis, gray-level co-occurrence and run-length matrix features were extracted and compared using Mann-Whitney U tests. ROC curves for each tumor were constructed, and AUCs were calculated.
Overall, sarcomatoid RCCs were larger than clear cell RCCs, measuring 77 ± 27 mm (mean ± SD) compared with 50 ± 29 mm (p = 0.003), respectively; however, there was no difference in tumor size between the tumors that were compared using texture analysis or subjective analysis (p = 0.06 and 0.03, respectively). From the subjective analysis, only peritumoral neovascularity (readers 1 and 2: 70% and 70% sarcomatoid RCCs vs 0% and 41.6% clear cell RCCs, respectively; p = 0.001) and the size of the peritumoral vessels (p < 0.001) differed between sarcomatoid RCCs and clear cell RCCs, and interobserver agreement was fair (κ = 0.38). Other subjective imaging features did not differ between the tumors (p > 0.005). There was greater run-length nonuniformity and greater gray-level nonuniformity in sarcomatoid RCCs than in clear cell RCCs (p = 0.03 and p = 0.04, respectively). The combined textural features identified sarcomatoid RCC with an AUC of 0.81 ± 0.08 (standard error) (p < 0.0001).
Large tumor size, the presence of peritumoral neovascularity, and larger peritumoral vessels are features that are more commonly associated with sarcomatoid RCCs than with clear cell RCCs. Sarcomatoid RCCs are also more heterogeneous by texture analysis than clear cell RCCs.