Differentiating papillary type I RCC from clear cell RCC and oncocytoma: application of whole-lesion volumetric ADC measurement

Abstract

Purpose: To determine whether objective volumetric whole-lesion apparent diffusion coefficient (ADC) distribution analysis improves upon the capabilities of conventional subjective small region-of-interest (ROI) ADC measurements for prediction of renal cell carcinoma (RCC) subtype.

Methods: This IRB-approved study retrospectively enrolled 55 patients (152 tumors). Diffusion-weighted imaging DWI was acquired at b values of 0, 250, and 800 s/mm² on a 1.5T system (Aera, Siemens Healthcare). Whole-lesion measurements were performed by a research fellow and reviewed by a fellowship-trained radiologist. Mean, median, skewness, kurtosis, and every 5th percentile ADCs were determined from the whole-lesion histogram. Linear mixed models that accounted for within-subject correlation of lesions were used to compare ADCs among RCC subtypes. Receiver-operating characteristic (ROC) curve analysis with optimal cutoff points from the Youden index was used to test the ability of ADCs to differentiate clear cell RCC (ccRCC), papillary RCC (pRCC), and oncocytoma subtypes.

Results: Whole-lesion ADC values were significantly different between pRCC and ccRCC, and between pRCC and oncocytoma, demonstrating strong ability to differentiate subtypes across the quantiles (both P < 0.001). Best percentile ROC analysis demonstrated AUC values of 95.2 for ccRCC vs. pRCC; 67.6 for oncocytoma vs. ccRCC; and 95.8 for oncocytoma vs. pRCC. Best percentile ROC analysis further indicated model sensitivities/specificities of 84.5%/93.1% for ccRCC vs. pRCC; 100.0%/10.3% for oncocytoma vs. ccRCC; and 88.5%/93.1% for oncocytoma vs. pRCC.

Conclusion: The objective methodology of whole-lesion volumetric ADC measurements maintains the sensitivity/specificity of conventional expert-based ROI analysis, provides information on lesion heterogeneity, and reduces observer bias.

Keywords: Apparent diffusion coefficients; Diffusion-weighted imaging; Renal cell carcinoma; Subtype differentiation.

Fig. 1.

ADC patient cohort workflow.

Fig. 2.

Box and Whisker Plots of ADC value distributions among the subtypes. As illustrated, papillary RCC and clear cell RCC, and papillary RCC and oncocytomas were significantly different. However, oncocytomas and clear cell RCC did not demonstrate significant differences in ADC values.

Fig. 3.

AUC distribution per quantile for each model demonstrating overall model performance and best-performing quantile.

Fig. 4.

ROC analysis plots with best percentiles, sensitivity, specificity, and area under the curve (AUC) for each subtype.

Fig. 5.

ADC map and histogram of ccRCC lesion.

Fig. 6.

ADC map and histogram of pRCC lesion.

Fig. 7.

ADC map and histogram of oncocytoma lesions.