Document Type : Original Article
Authors
1
Department of Computer Engineering, Shahr-e-Qods branch, Islamic Azad Unversity, Tehran, Iran
2
Department of Computer Engineering, Shahr-e-Qods branch, Islamic Azad University, Tehran, Iran
10.48301/kssa.2023.404913.2625
Abstract
In recent years, there has been a significant increase in the use of deep learning models for analyzing medical images. This research introduces a novel approach for diagnosing cervical cancer using a deep learning ensemble model based on belief theory and microscopic cell images. The proposed model, called DEBN, utilizes a deep belief network and an ensemble classification model to enhance the accuracy of predictions. To address the uncertainty involved in cell type classification, this method incorporates a belief network and Dempster's combination rule to combine evidence effectively. Additionally, a new ensemble classification model is introduced to classify rejected samples that exhibit high uncertainty in the Dempster method. To improve the performance of the model, five classifiers including Support Vector Machine, K-Nearest Neighbor, decision trees, Naïve Bayes, and a fuzzy classifier are used to classify samples. The experimental results on the Herlev dataset and the SIPaKMeD dataset demonstrate the superiority of the proposed DEBN model. On the Herlev dataset, the DEBN model achieves an accuracy of 99.93%, specificity of 98.53%, sensitivity of 98.83%, and an AUC of 99.94%. Similarly, on the SIPaKMeD dataset, the model achieves an accuracy of 97.2%, specificity of 98.79%, sensitivity of 98.70%, and an AUC of 99.89%. These results surpass those obtained by traditional deep learning methods, showcasing the promising potential of early detection of cervical cancer using the Deep Ensemble Belief Model.
Keywords
Main Subjects
)