A novel variant of deep convolutional neural network for classification of ovarian tumors using CT images

• Published in: Computers & electrical engineering 2023-08, Vol.109, p.108758, Article 108758
• Main Authors: Kodipalli, Ashwini, Devi, Susheela V, Dasar, Santosh, Ismail, Taha
• Format: Article
• Language: English
• Subjects: Big data; Cloud computing; Computational intelligent framework; Deep neural networks; Google cloud; ILSVRC; Ovarian tumors
• ISSN: 0045-7906; 1879-0755
• DOI: 10.1016/j.compeleceng.2023.108758

•Though computerized tomography can contain minute and subtle details of ovarian tissue which would be predominantly used to provide discrimination in classification as benign or malignant, yet there are substantially limited studies.•This research work mainly focused on comparing the various variants of CNN architecture for the task of classification of tumors into benign and malignant using CT scan dataset.•To take care of the limitations of the existing work, the current research work is designed to include computerized tomography (CT) scanned ovarian images to detect and classify tumours as benign or malignant using newly state-of-the-art deep learning model and to compare the results with the classification carried out by expert radiologists. Deep Learning models have shown tremendously impressive performance on image classification tasks. In the medical imaging domain, progress has been made in obtaining high-quality data for analysis and using state-of-the- art artificial intelligence algorithms for solving complex problems and providing answers to key questions using data. One such problem that is of crucial importance and interest to medical researchers is to classify tumors into two categories benign and malignant. This research work focuses on proposing a novel variation of CNN architecture and a comparison of the performances of state-of-the-art ImageNet Large Scale Visual Recognition Challenge (ILSVRC) winning architectures for the task of classifying ovarian tumors by training and evaluating images on a dataset of ovarian CT scan images with the help of cloud services such as Google Cloud Platform. The proposed architecture has attained an accuracy of 97.53% and outperformed the existing CNN variants.