Abstracts

RETURN TO ABSTRACT LISTING


E1050. Evaluation of a Texture Analysis Based on MRI for Differentiation of Focal-Type Autoimmune Pancreatitis from Pancreatic Carcinoma
Authors
  1. Megumi Shiraishi; The Jikei University School of Medicine
  2. Takao Igarashi; The Jikei University School of Medicine
  3. Kazuyoshi Ohki; The Jikei University School of Medicine
  4. Hiroya Ojiri; The Jikei University School of Medicine
Objective:
Imaging-based differentiation of focal-type autoimmune pancreatitis (f-AIP) from pancreatic ductal adenocarcinoma (PDAC) is difficult. However, f-AIP is associated with low apparent diffusion coefficient (ADC) [1-3]. The purpose of our study was to evaluate the usefulness of 3D volume-of-interest (VOI)- and ADC-based texture analysis to differentiate f-AIP from PDAC.

Materials and Methods:
Our IRB approved study included 29 patients with f-AIP and 84 patients with pathologically proven PDAC who underwent magnetic resonance imaging (MRI). Two readers retrospectively evaluated 3D VOI-based (3D-ADCmean) mean ADC values and 288 automatically calculated texture parameters (16 texture features×18 filters) from all MR images. Categorical and continuous variables (patient background, ADC, texture features) were compared using chi-square and Mann-Whitney U tests, respectively. Optimal cut-off values were calculated using areas under receiver operating characteristic curves (AUCs) and Youden index analysis. Inter-reader reliability was estimated by intra class correlation coefficient (ICC). Multivariate logistic regression analysis was performed to identify the independent factors for differentiation between f-AIP and PDAC.

Results:
There was a significant male predominance among patients with f-AIP (p = 0.03). The duct-penetrating sign, capsule-like rim, and multiple MPD strictures were significantly more frequently present in the f-AIP than in the PDAC group (p<0.05). 3D-ADCmean was significantly lower in f-AIP (1.16–1.19 × 10–3 mm2/s) vs. PDAC (1.41–1.43 × 10–3 mm2/s), with good inter-reader reliability (ICC=0.873). Among texture features, energy with exponential filtration yielded the highest AUC (Reader 1, 76.7%; Reader 2, 83.8%), with moderate inter-reader reliability (ICC=0.700). On multivariate analysis, exponential-energy (odds ratio [OR], 3.699; 95% confidence interval [CI], 1.082–12.644; p<0.05) and 3D-ADCmean (OR, 7.151; 95% CI, 1.006–50.812; p<0.05) were significant predictors for differentiation between f-AIP and PDAC.

Conclusion:
Exponential-energy and 3D-ADCmean may improve MRI-based differentiation between f-AIP and PDAC.