Future Generation Computer Systems 102 (2020) 643–649 Contents lists available at ScienceDirect Future Generation Computer Systems journal homepage: www.elsevier.com/locate/fgcs Cervical cancer classification using convolutional neural networks and extreme learning machines Ahmed Ghoneim a,∗ , Ghulam Muhammad b , M. Shamim Hossain a a Department of Software Engineering, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia b Department of Computer Engineering, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia article info Article history: Received 2 April 2019 Received in revised form 28 August 2019 Accepted 9 September 2019 Available online 13 September 2019 abstract Cervical cancer is one of the main reasons of death from cancer in women. The complication of this cancer can be limited if it is diagnosed and treated at an early stage. In this paper, we propose a cervical cancer cell detection and classification system based on convolutional neural networks (CNNs). The cell images are fed into a CNNs model to extract deep-learned features. Then, an extreme learning machine (ELM)-based classifier classifies the input images. CNNs model is used via transfer learning and fine tuning. Alternatives to the ELM, multi-layer perceptron (MLP) and autoencoder (AE)-based classifiers are also investigated. Experiments are performed using the Herlev database. The proposed CNN-ELM-based system achieved 99.5% accuracy in the detection problem (2-class) and 91.2% in the classification problem (7-class). © 2019 Published by Elsevier B.V. 1. Introduction The cervix of a human is covered by a thin layer of tissues consisting of cells. If a cell is changed into a malignant cell that can grow and divide rapidly and becomes a tumor, we call this situation the cervical cancer. This cancer can be treated if it is detected at an early stage. The diagnosis is normally done by a screening process and a biopsy. Image processing techniques can be applied to find the spread of the cancer. Cervical cancer is the fourth-most common cause of death from cancer in women [1]. Medical image processing and intelligent systems play a role in the analysis of the malignant cells. With the development of new techniques, they become cost-effective and less time- consuming. They are now becoming popular over conventional methods such as Pap Smear, Colposcopy, and Cervicography. These techniques are unbiased to human experience; however, we want to stress that they cannot replace the subjective (expert doctor) evaluation, but can assist them to a great degree. State-of-the-art machine learning techniques and wireless communication technologies have enabled us to develop a com- plete medical diagnosis system that can operate in real-time, accurately, and without human interaction. Yet, there are many issues that need to be solved, for example, packet loss during transmission, high bandwidth requirement for medical video data transfer, and a robust algorithm to deal with many varia- tions in data. To address some of this issues, edge-based cloud ∗ Corresponding author. E-mail address: ghoneim@ksu.edu.sa (A. Ghoneim). computing was proposed for voice pathology detection [2,3], Internet of Things (IoTs) and cloud-based framework was realized in [4], deep learning for emotion recognition was proposed in [5], edge-based communication was introduced in [6], and a disease monitoring system was proposed in [7]. Computer-aided systems in cancerous cell detection have been used in the literature for quite a few times. In breast cancer recognition, different feature extraction methods such as local binary pattern, histogram of gradient orientation, and Laplacian Gaussian filter were used [8–10]. Local texture analysis was used to diagnose pulmonary nodules in [11]. To analyze dermoscopy images for skin cancer, directional filters and color component features were used in [12,13]. A method for voice pathology detection using different input modalities were proposed in [14, 15]. Recently, deep learning has brought a big improvement in accuracy in many applications. Due to its high accuracies in many areas, it has become the state-of-the-art machine learning technique. A good survey on various cancerous cells detection using deep learning can be found in [16]. Deep learning was successfully used in EEG pathology detection [17,18], environ- ment classification [19], lung nodule detection [20], breast cancer detection [21], skin cancer detection [22], medical image anal- ysis [23], audio–visual emotion recognition [24] and diseases prediction [25,26]. With the increase in many types of sensors, processing of Big Data has added an extra dimension to the deep learning. Big Data has successfully handled in several medical related applications [27,28]. Due to the success of deep learning in many medical appli- cations, in this paper, we propose a deep learning-based system https://doi.org/10.1016/j.future.2019.09.015 0167-739X/© 2019 Published by Elsevier B.V.