A novel pathological stroke classification system using NSST and WLEPCA

Abstract

Stroke is a type of cerebrovascular disease, and it is one of the leading cause of death, with over six million deaths recorded annually. In this paper, a novel scheme for an accurate multi-class stroke disease classification named Pathological Stroke Classification System (PSCS) is introduced to classify stroke disease into six classes. Features are extracted using Nonsubsampled Shearlet Transform (NSST), which decomposes the fused image into the low-frequency band and k-bands of high frequency. The low-frequency band is further analyzed using a new scheme of feature reduction and selection using weighted local energy based principal component analysis (WLEPCA). Different subsets of principal vectors are applied to three decision models, k-Nearest Neighbors (KNN), random forest (RF), and Support Vector Machine (SVM). The RF-based classifier performed better than SVM and k-NN and achieved an accuracy of 96.10%. The proposed PSCS showed a promising result in stroke classification can be considered as a reliable and robust diagnostic tool for medical practitioners.