Abstract Archives of the RSNA, 2014
Classification of Renal Masses Using CT Texture Analysis: Generation of A Predictive Model on the Basis of Quantitative Spatial Frequency Measurements and Random Forest Modeling
Presented on December 2, 2014
Presented as part of SSG07: Informatics (3D, Quantitative and Advanced Visualization)
Siva P. Raman MD, Presenter: Nothing to Disclose
Yifei Chen BS, Abstract Co-Author: Nothing to Disclose
James L. Schroeder MD,PhD, Abstract Co-Author: Nothing to Disclose
Peng Huang, Abstract Co-Author: Nothing to Disclose
Elliot K. Fishman MD, Abstract Co-Author: Research support, Siemens AG
Advisory Board, Siemens AG
Research support, General Electric Company
Advisory Board, General Electric Company
Co-founder, HipGraphics, Inc
CT texture analysis (CTTA) allows the quantification of lesion heterogeneity based on the distribution of pixel intensities within a region of interest. This study investigates the ability of CTTA to distinguish between several different common renal masses, and seeks to develop a ‘random forest’ predictive model allowing the differentiation of these lesion types.
Following IRB approval, CTTA software (TexRAD Ltd.) was used to retrospectively analyze 20 clear cell renal cell carcinomas, 20 papillary RCCs, 20 oncocytomas, and 20 Bosniak I renal cysts. Regions of interest were drawn around each mass on multiple slices, and each lesion was analyzed using arterial, venous, and delayed phase images on renal mass protocol CT scans performed with uniform technique. Analysis was performed on both unfiltered images and spatial band-pass filtered images to quantitatively assess heterogeneity. Random forest method was used to construct a predictive model to classify lesions, and separate models were constructed using either one phase in isolation or all three contrast phases in conjunction. The model was then externally validated on a separate set of 19 cases that were not used in the generation of the original random forest model.
The random forest model was able to successfully distinguish the four lesion types, and when utilizing all 3 contrast phases in conjunction, the model correctly categorized oncocytomas in 89% of cases (sensitivity 89%, specificity 99%), clear cell RCCs in 91% of cases (sensitivity 91%, specificity 97%), cysts in 100% of cases (sensitivity 100%, specificity 100%), and papillary RCCs in 100% of cases (sensitivity 100%, specificity 98%). Models utilizing a single contrast phase (arterial, venous, or delayed) in isolation were less accurate, with the model based only on the arterial phase images performing the best. When tested on a separate/validation set of 19 cases, the model was correctly able to categorize all 19 cases.
CTTA, in conjunction with random forest modeling, demonstrates great promise as a tool for correctly characterizing lesion types, and was able to classify four common types of renal masses with a high degree of accuracy.
CT texture analysis allowed accurate characterization of a small subset of common renal masses, suggesting its promise as a quantitative imaging tool that may augment our ability to predict lesion histology.
Classification of Renal Masses Using CT Texture Analysis: Generation of A Predictive Model on the Basis of Quantitative Spatial Frequency Measurements and Random Forest Modeling. Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14001334.html