An Improved Fuzzy Deep Learning (IFDL) model for managing uncertainty in classification of pap-smear cell images

Abstract

Applications of deep learning models for medical image analysis have been concentrated in the recent years. An automatic detection system to detect the class of Pap smear cell and cervical cancer is a challenging problem due to time consuming and erroneous process of the detection for technicians. This study presents an improved Deep Convolutional Neural Network (DCNN) for analysis of Pap smear images for early detection of cervical cancer. The proposed model addresses the issue of classification of samples with similar probability in classification layer of a DCNN. To address this challenge, an Improved Fuzzy Deep learning (IFDL) model has been proposed by taking advantages of Deep Belief Network using Dempster combination rule, and Fuzzy weighting system, to manage uncertainty of similar classes in the classification layer. In this method, a new layer by Belief Networks using Dempster combinational rule, aggregates the evidences to handle uncertainty of assigning correct class, between different classes. To address the issue of the object rejection in Belief network, a fuzzy weighting system has been proposed. The experimental results for two classes problem and seven classes problem on Herlev cell image dataset, show the superiority of the proposed model. This model with an accuracy of 99.20% outperforms counterpart methods and is promising for early detection of the cervical cancer.

Keywords

• Deep learning
• Uncertainty handling
• Dempster-Shafer theory of evidence
• Belief network
• Cell image classification, Fuzzy systems