Abstract A83: Nuclear morphometry differentiates chronic pancreatitis, IPMN, and pancreatic carcinoma

Abstract

Abstract Background: It can be difficult to distinguish between chronic pancreatitis (CP), IPMN, and pancreatic carcinoma (PC) on tissue biopsy. Nuclear morphometry can measure up to 93 unique nuclear features based on standard histopathology. The goal of this work is to build novel, objective, and accurate prediction tools, based on nuclear morphometric signatures in high resolution images of nuclei of histologic sections, for classifying pancreatic tissues into three distinct groups. Materials & Methods: 44 patients who underwent pancreatic resections were identified. 12 cases of CP, 16 cases of IPMN, and 16 cases of PC were utilized in this pilot study. 180 ± 22 nuclei from each lesion were imaged with high resolution microscopy. Clincodemographic data was obtained retrospectively from the medical record. Statistically significant nuclear features were determined by a fully automated penalized multinomial regression algorithm in order to determine a multi-class classifier and simultaneously identify important nuclear features. The LASSO penalty function, and associated regularization parameter, is adaptively chosen by cross validation to prevent over-fitting. In order to test the veracity of the automated algorithm, we randomly removed 25% of the cases as a training set and utilized the remaining cases as a test set; this was repeated 10 times. Results: The average age was 64 ± 15 years, with patients in the CP being slightly younger; 63% were male. Median follow-up time was 3 years in the CP group, 3 years in the IPMN group, and 5 years in PC group. The method described automatically identified 6 unique and statistically significant nuclear features (corrected overall P < 0.005); these features related to nuclear heterogeneity and chromatin shapes. Cases assigned to 1 of the 3 groups were correctly assigned with an overall accuracy of 84.5%. Interestingly, cases of incorrect classification had larger proportion of nuclei from all three groups suggesting a mixed morphologic lesion. Conclusions: Nuclear morphometry classifies pancreatic lesions into CP, IPMN, and PC with 84.5% accuracy using a fully automated algorithm to determine statistically significant and unique nuclear features. Since the incorrectly classified lesions had a larger proportion of mixed nuclei, diagnostic uncertainty may be determined in a quantitative manner allowing for a confidence probability estimation of whether a given lesions should be classified as a CP, IPMN, or PC. Further studies will validate these results in a resected cohort as well as a cohort based on biopsied specimens alone. Citation Format: Evan S. Glazer, Hao Zhang, Kimberly A. Hill, Charmi Patel, Stephanie T. Kha, Peter H. Bartels, Michael L. Yozwiak, Hubert G. Bartels, Joseph C. Watkins, David S. Alberts, Robert S. Krouse. Nuclear morphometry differentiates chronic pancreatitis, IPMN, and pancreatic carcinoma. [abstract]. In: Proceedings of the AACR Special Conference on Pancreatic Cancer: Innovations in Research and Treatment; May 18-21, 2014; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2015;75(13 Suppl):Abstract nr A83.