Deep learning IoT system for online stroke detection in skull computed tomography images

• Published in: Computer networks (Amsterdam, Netherlands : 1999) Vol. 152; pp. 25 - 39
• Main Authors: Dourado Jr, Carlos M.J.M., da Silva, Suane Pires P., da Nóbrega, Raul Victor M., da S. Barros, Antonio Carlos, Filho, Pedro P. Rebouças, de Albuquerque, Victor Hugo C.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 07.04.2019; Elsevier Sequoia S.A
• Subjects: Artificial neural networks; Bayesian analysis; Brain; Classifiers; Computation; Computed tomography; Deep learning; Diagnostic systems; Image classification; Image detection; Intelligent IoT; Internet of Things; Machine learning; Medical imaging; Multilayer perceptrons; On-line systems; Recall; Remote monitoring; Stroke; Support vector machines; Testing time; Tomography; Intelligent IoT; Computed tomography; Stroke
• ISSN: 1389-1286; 1872-7069
• DOI: 10.1016/j.comnet.2019.01.019

Cerebral vascular accidents (CVA) affect about 16 million people worldwide annually. CVA, also know as stroke, is a serious global health problem, and can cause significant physical limitations to those affected. Computed tomography is the most appropriate procedure to diagnose and evaluate the dimensions and magnitude of a stroke. Thus, in this article we present an Internet of Things (IoT) framework for the classification of stroke from CT images applying Convolutional Neural Networks (CNN) in order to identifying a healthy brain, an ischemic stroke or a hemorrhagic stroke. Following the Transfer Learning concept CNN was combined with different consolidated Machine Learning methods such as Bayesian Classifier, Multilayer Perceptron, k-Nearest Neighbor, Random Forest and Support Vector Machines. Our approach contributes to the automation of the diagnostic process by a competent method that is able to obtain information imperceptible to the human eye, and thus it contributes to a more precise diagnosis. In addition, with the advent of IoT, a highly efficient and flexible new instrument emerges to address issues related to health care services and specifically in our approach can provide remote diagnoses and monitoring of patients. The approach was validated by analyzing the parameters Accuracy, F1-Score, Recall, Precision and processing time. The results showed that CNN obtained 100% Accuracy, F1-Score, Recall and Precision in combination with most of the classifiers tested. The shortest training and test times were 0.015 s and 0.001 s, respectively, both in combination with the Bayesian Classifier. Thus, our proposed approach demonstrates efficiency and reliability to detect strokes.