Two-stage high-frequency ultrasonic characterization of ex vivo renal tumor microstructure

Pre-operative classification of renal tumor subtypes remains challenging. Renal mass biopsies are a solution, but are invasive and currently underutilized in clinical practice. This study evaluated the feasibility of a two-stage, high-frequency ultrasonic characterization of ex vivo renal tumor subtypes to non-invasively assess the tissue microstructure and subsequently classify the tumor histologic subtype.

Ex vivo renal tumors resected by partial or radical nephrectomy were scanned immediately after surgery using a high-frequency 29 MHz ExactVu (Exact Imaging, ON, Canada) ultrasound device. From an initial cohort of 153 unique tumor samples, 91 ultrasound cine loops (one per sample, including sample pruning) were evaluated to identify common radiological (or visual) features among the different histologic subtypes. In this study, we considered clear cell renal cell carcinoma (RCC), papillary RCC, chromophobe RCC, oncocytoma and angiomyolipoma histologic subtypes. In a subsequent cohort of 61 samples selected for quantitative ultrasound (QUS)-based analysis (based on the availability of radiofrequency echo data), parameters computed using the ultrasound radiofrequency echo data were used to develop a linear, benign-versus-malignant subtype classifier. The homodyned K-distribution structure parameter and the effective acoustic concentration were used.

Six ultrasound image features describing the histologic subtype appearances were observed and cataloged, with the features comprising well-defined borders, rounded shape, homogeneous texture, cystic components, septations, and punctate foci. Classification of 36 samples using these imaging features alone demonstrated an accuracy of only 30.6%, hereby confirming that visual pattern recognition by the human eye is insufficient for accurate histologic subtype classification and thus motivated the quantitative ultrasound-based approach developed in stage 2. In contrast, the QUS-based linear classifier achieved 82% accuracy, 82% sensitivity and 80% specificity using 61 samples.

We demonstrated the feasibility of QUS-based classification of renal tumor histologic subtypes, as well as a catalog of six correlated ultrasound image features. This non-invasive, ex vivo approach shows promise to guide diagnosis and reduce unnecessary surgeries in urologic cancer. Future work includes validation of the built QUS-based classifier on a larger test cohort, followed by in vivo feasibility studies conducted prior to nephrectomy.