Quantitative imaging to evaluate malignant potential of IPMNs

// Alexander N. Hanania 1,2 , Leonidas E. Bantis 3 , Ziding Feng 3 , Huamin Wang 4 , Eric P. Tamm 5 , Matthew H. Katz 6 , Anirban Maitra 4 and Eugene J. Koay 2 1 University of Texas Medical School, Houston, TX, USA 2 Department of Radiation Oncology, University of Texas MD Anderson Cancer Center, Houston, TX, USA 3 Department of Biostatistics, University of Texas MD Anderson Cancer Center, Houston, TX, USA 4 Department of Pathology, University of Texas MD Anderson Cancer Center, Houston, TX, USA 5 Department of Diagnostic Radiology, University of Texas MD Anderson Cancer Center, Houston, TX, USA 6 Department of Surgical Oncology, University of Texas MD Anderson Cancer Center, Houston, TX, USA Correspondence to: Eugene J. Koay, email: // Keywords : IPMN, quantitative imaging, radiomics, fukuoka, pancreatic cyst Received : July 06, 2016 Accepted : August 18, 2016 Published :August 31, 2016 Abstract Objective: To investigate using quantitative imaging to assess the malignant potential of intraductal papillary mucinous neoplasms (IPMNs) in the pancreas. Background: Pancreatic cysts are identified in over 2% of the population and a subset of these, including intraductal papillary mucinous neoplasms (IPMNs), represent pre-malignant lesions. Unfortunately, clinicians cannot accurately predict which of these lesions are likely to progress to pancreatic ductal adenocarcinoma (PDAC). Methods: We investigated 360 imaging features within the domains of intensity, texture and shape using pancreatic protocol CT images in 53 patients diagnosed with IPMN (34 “high-grade” HG and 19 “low-grade” LG) who subsequently underwent surgical resection. We evaluated the performance of these features as well as the Fukuoka criteria for pancreatic cyst resection. Results: In our cohort, the Fukuoka criteria had a false positive rate of 36%. We identified 14 imaging biomarkers within Gray-Level Co-Occurrence Matrix (GLCM) that predicted histopathological grade within cyst contours. The most predictive marker differentiated LG and HG lesions with an area under the curve (AUC) of .82 at a sensitivity of 85% and specificity of 68%. Using a cross-validated design, the best logistic regression yielded an AUC of 0.96 (σ = .05) at a sensitivity of 97% and specificity of 88%. Based on the principal component analysis, HG IPMNs demonstrated a pattern of separation from LG IPMNs. Conclusions: HG IPMNs appear to have distinct imaging properties. Further validation of these findings may address a major clinical need in this population by identifying those most likely to benefit from surgical resection.