Detection of Stroke in Brain CT with Fused Deep-Features and Hummingbird Optimizer
Swaetha Ramadasan, Manigandan Ramadasan, K. Vijayakumar, Gangadharam Balaji
Abstract
Brain is one of the sensitive organs and the abnormality in the brain is a medical emergency. Hemorrhagic Stroke (HS) is one of the harsh brain condition and appropriate screening and treatment is necessary to reduce its harshness. Medical imaging supported brain screening is a common clinical practice to examine the HS. This work proposes Deep-Learning Technique (DT) to detect HS from the CT scan slices. The stages in this scheme includes; (i) CT slice collection and resizing, (ii) feature extraction using DT, (iii) Hummingbird-Algorithm based feature selection and serial features concatenation and (iv) classification and verification of the performance. This research considered N=2000 images (1000 healthy and 1000 stroke) along with the skull section. The performance of EfficientNet and DenseNet variants are considered to examine the CT slices. The experimental outcome of this study confirms that the fused optimal features helps to attain an accuracy of >96%. In future, this performance of this system can be improved by considering ensemble of features.