Prediction of Tumor-Infiltrating CD20+ B-Cells in Patients with Pancreatic Ductal Adenocarcinoma Using a Multilayer Perceptron Network Classifier Based on Non-contrast MRI.

Abstract

RATIONALE AND OBJECTIVES:

Conventional chemotherapy has limited benefit in pancreatic ductal adenocarcinoma (PDAC), necessitating identification of novel therapeutic targets. Radiomics may enable non-invasive prediction of CD20 expression, a hypothesized therapeutic target in PDAC. To develop a machine learning classifier based on noncontrast magnetic resonance imaging for predicting CD20 expression in PDAC.

MATERIALS AND METHODS:

Retrospective study was conducted on preoperative noncontrast magnetic resonance imaging of 156 patients with pathologically confirmed PDAC from January 2017 to April 2018. For each patient, 1409 radiomics features were selected using minimum absolute contraction and selective operator logistic regression algorithms. CD20 expression was quantified using immunohistochemistry. A multilayer perceptron network classifier was developed using the training and validation set.

RESULTS:

A log-rank test showed that the CD20-high group (22.37 months, 95% CI: 19.10-25.65) had significantly longer survival than the CD20-low group (14.9 months, 95% CI: 10.96-18.84). The predictive model showed good differentiation in training (area under the curve [AUC], 0.79) and validation (AUC, 0.79) sets. Sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were 73.21%, 75.47%, 0.74, 0.76, and 0.73, respectively, for the training set and 69.23%, 80.95%, 0.74, 0.82, and 0.68, respectively, for the validation set.

CONCLUSION:

Multilayer perceptron classifier based on noncontrast magnetic resonance imaging scanning can predict the level of CD20 expression in PDAC patients.

• Authors: Qi Li, Jieyu Yu, Hao Zhang, Yinghao Meng, Yan Fang Liu, Hui Jiang, Mengmeng Zhu, Na Li, Jian Zhou, Fang Liu, Xu Fang, Jing Li, Xiaochen Feng, Jianping Lu, Chengwei Shao, Yun Bian
• Journal: Academic radiology (Acad Radiol) Vol. 29 Issue 9 Pg. e167-e177 (09 2022) ISSN: 1878-4046 [Electronic] United States
• PMID: 34922828 (Publication Type: Journal Article, Research Support, Non-U.S. Gov't)
• Copyright: Copyright © 2021 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.

Topics

• Carcinoma, Pancreatic Ductal (diagnostic imaging, pathology)
• Humans
• Magnetic Resonance Imaging (methods)
• Neural Networks, Computer
• Pancreatic Neoplasms (diagnostic imaging, pathology)
• Retrospective Studies