Deep learning-based approach for prediction of brain stroke from MR images for IoT in healthcare
Manu Gupta, Pessani Meghana, Kancherla Harshitha Reddy
Abstract
<p class="CM12">This study develops a technique to predict brain strokes using magnetic resonance imaging (MRI). Worldwide, brain stroke is a leading factor in death and long-term impairment. The impact of stroke on the life of survivors is substantial, often resulting in disability. Stroke analysis performed manually takes a lot of time and is subject to intra- and inter-operator variability. Consequently, this work aims to create a computer-based system for the prediction of stroke utilizing deep learning techniques, which help in timely diagnosis. The MRI images are preferred as it provides images of good contrast and no ionizing radiations are used in this imaging method. The deep learning methods included in this proposed work are DenseNet-121, Xception, LeNet, ResNet-50 and VGG-16. The DenseNet-121 classifier outperformed other classifiers and achieved acccuracy of 96%. The outcomes of the proposed approach for stroke prediction in IOT healthcare systems show that improved performance is attained using deep learning methods.</p>