Pd14-07 Differentiation Between Clear Cell Renal Cell Carcinomas and Oncocytomas Using Texture Analysis of Ct Images

Abstract

imaging technique used for the rapid histological diagnosis of renal cell carcinoma on percutaneous biopsy.METHODS: After institutional board approval, 50 total percutaneous biopsies were obtained from 20 patients undergoing either partial or radical nephrectomy from resected renal tissue after specimen removal. Biopsy specimens were sprayed with a dual topical fluorescent stain, and VR-SIM images obtained using a 4.2 megapixel, high-speed CMOS camera with fast ferroelectric spatial light modulators. Biopsy specimens were subsequently fixed and analyzed using standard hematoxylin and eosin (H&E) histopathologic methods. One blinded pathologist reviewed all VR-SIM images and H&E specimens. The presence of cancer was correlated between the two modalities and final standard pathology result.RESULTS: Of the 50 biopsy specimens, fourteen were determined to be malignant on H&E, while 36 represented benign renal tissue. Thirteen of the 14 malignant specimens were diagnosed by VR-SIM, with 1 specimen being assessed as indeterminant. Sensitivity and specificity of VR-SIM were 93% and 100%, respectively.CONCLUSIONS: VR-SIM is a promising and practical method for point of care renal biopsy assessment. This may translate into the ability to rapidly assess sufficiency of biopsy specimens and enable immediate diagnostic capability to aid in counseling patients regarding management options.

A Decision-Support Tool for Renal Mass Classification

“…were analyzed. We derived features using various techniques ranging from simple histogram analysis to more advanced techniques such as gray-level co-occurrence matrix (GLCM) and gray-level difference matrix (GLDM) [8,9,25,26]. The techniques (summarized in Table 3) have been described in greater detail in previous studies [27][28][29][30][31][32].…”

“…Factors beyond tumor size and contrast enhancement, such as tumor texture and shape, have been associated with tumor behavior but are currently not commonly used in clinical decision-making. Radiomics, the high-throughput extraction of such quantifiable image features, is a natural next step [8,9].…”

“…More worryingly, as the network becomes deeper, the model identifies increasingly complex combinations of features that depend on previous layers, making the model difficult to interpret or explain. Instead of letting a deep network identify potentially unintelligible features, we use radiomic features that have proven clinical relevance[8,9,25,26] as well as explainability. Using RFGB, we identify combinations of these features and discover relationships.…”