In today’s Society of Urologic Oncology (SUO) meeting at the 2016 AUA, Dr. Robert Uzzo raised the question as to whether we are ready for the “radiographic biopsy” for the diagnosis of kidney cancer. In line with the trend of scientific questions being answered utilizing large data sets, radiologic comparative effectiveness studies are well positioned to capitalize on “big data;” thousands of images are acquired in both MRI and CT, yet only a small percentage are actually used by the astute clinician.
A sizable amount of data is thus left “on the table,” and it prompts discussion and specification of what quantifiable features can be extracted and acted upon. Analysis of such features, dubbed “radiomics,” can include imaging signal features (density on CT, intensity on MRI, echogenicity on US), physical features (size, volume, shape, lobularity, architecture), radiologic/anatomical/spatial features (enhancement, tumor/organ relation, texture and cluster analyses), and other features such as molecular imaging and epigenetics.
The practical importance of radiomics is emphasized by our ability to measure the biologic risk of a renal mass without actually obtaining tissue. Currently, we use gestalt in addition to clinically validated models to assess risk. The ultimate goal may be liquid biopsy. Medium points on this continuum include percutaneous biopsy and “radiographic biopsy.” Contrast CT scan can differentiate clear cell, oncocytoma, papillary, chromophobe, and fat-poor AML simply with the assessment of pattern (hetero vs. homogeneous), enhancement, neovascularity, contour, and washout. Differentiation in mean attenuation by multiphase detectors can further classify tumors by histologic subtype. By optimizing the threshold of peak enhancement, we can differentiate clear cell from papillary, chromophobe, and oncocytoma with accuracy ranging 77%-85%.
Similarly, comparison of relative enhancement in the contrast vs. non-contrast phase can provide 95% sensitivity and 92% specificity. Computerized quantification of texture, heterogeneity, and tumor grade (by identification of quantifiable differences in microvascular density) can further enhance diagnostic accuracy. MRI, which complements CT, can further characterize lesions with the use of T1 and T2 imaging, gadolinium, fat suppression, subtraction imaging, and diffusion-weighted imaging. Clear cell RCC tend to have increased enhancement, increased intensity on T2, and >25% on I/O phase. Chromophobe RCCs tend to have slightly less enhancement, intermediate T2 signal and I/O phases. Papillary RCCs have low enhancement and T2 signal intensity. Oncocytomas have high enhancement, intermediate T2 signal intensity and I/O phases. Finally, lipid poor AML has mild enhancement, low T2 intensity, and >25% on I/O phase. Diffusion-weighted imaging has an 86% sensitivity and 78% specificity for detection of benign vs. malignant disease. While it cannot distinguish clear cell from other histologies, it may be useful in differentiating low vs. high grade. Nuclear imaging is another modality that can facilitate radiographic biopsy. Girentuximab (cG250) is a chimeric monoclonal antibody that binds to CAIX, which is upregulated in more than 95% of clear cell cases. It has a sensitivity of 86% and specificity of 87%, with a PPV of 95%.
One of the more exciting developments in imaging is 99Tc-sestamibi san, a mitochondrial agent that has demonstrated 87.5% and 95% specificity. The initial study had 50 patients, in which there was one false negative and two false positive results. Dr. Uzzo concluded that as the field of radiomics is further developed, we will be better able to utilize the full armamentarium of imaging modalities at our disposal, and will ultimately be able to harness more out of the data we produce and leave less on the table. With radiomics, we can optimize big data and make better choices.
Reported by: Dr. Nikhil Waingankar, Fox Chase Cancer Center, Philadelphia, PA. for the AUA 2016 Annual Meeting in San Diego, CA.