Digital Imaging Tools for Differentiating between Type 1 and Type 2 Papillary Renal Cell Carcinoma
Jennifer A Hipp, Jason D Hipp, Kunju P Lakshmi, Ulysses J Balis. University of Michigan Health Systems, Ann Arbor, MI
Background: The recent availability of digital whole slide data sets has created new opportunities for pathologists to perform numerical and quantitative assessment of histologic features. One promising pattern matching algorithm, Spatially-Invariant Vector Quantization (SIVQ), has already exhibited broad utility in the detection of subtle architectural and nuclear features, making a compelling case for the exploration of its utility to differentiate between type 1 and type 2 Papillary Renal Cell Carcinoma (PRCC). Successful automated distinction of these two diagnostic entities would have immediate utility for the diagnosing pathologist.
Design: Digital whole slide images were obtained from PRCC type I and II cases. Use of a region-of-interest extraction tool, dCore, followed by use of an image aggregation tool, ImageMicroArray Maker, allowed for generation of a montage of diverse examples of each sub-type, as a combined monolithic image. Type 1 and type 2 image arrays served as a vehicle for expedited analysis of candidate features. Ring vectors were chosen for their ability to exhibit high affinity for: 1) the interface between dark blue nuclei and small nucleoli (vector 1, type 1 RRCC) and 2) areas containing nucleoli with white clearing and a blue nuclear rim (vector 2, type 2 PRCC). SIVQ analysis with these vectors was then performed on the representative region for each case, for confirmation of each ring vector's discriminant efficacy.
Results: From the initial validation region, vector 1 (type 1 PRCC) identified 47/54 nuclei (sensitivity 87%) as well as 3/35 (specificity 91%) of type 2 PRCC nuclei. Similarly, vector 2 (type 2 PRCC) identified 25/35 nuclei (sensitivity 87%) and 2/54 type 1 PRCC nuclei (specificity 96%). Confirming the above chosen vectors in the representative regions, the sensitivity and specificity was found to be 95% & 95% and 87% & 84% for vector 1 and 2, respectively.
Conclusions: SIVQ allows for efficient identification of both type 1 and type 2 PRCC nuclei, with high sensitivity and specificity. We attribute this slight differential performance of the two vectors to the observation that type 1 PRCC has increased homogeneity, leading to increased pattern matching performance (higher sensitivity) as compared to type 2 PRCC type, which exhibited large irregular nuclei with prominent nucleoli (decreasing homogeneity). We anticipate that such feature detection tools will facilitate the creation of turnkey decision support systems for the pathologist pitted with the distinction between these two diagnostically important morphological variants.
Category: Genitourinary (including renal tumors)
Wednesday, March 21, 2012 9:30 AM
Poster Session V # 107, Wednesday Morning