Stroke Disease and its Types Prediction using Images

Abstract

Stroke, a leading cause of long-term disability and mortality worldwide, arises from acute cerebrovascular events that disrupt cerebral blood flow. This neurological disorder necessitates rapid diagnostic imaging and timely intervention to minimize irreversible brain damage and improve patient outcomes. The disease prediction requires a structured image-based analysis approach that can identify the presence of stroke and provide further interpretation of its progression. This study presents a hierarchical deep learning framework for stroke disease prediction from brain scan images. The proposed system performs prediction in three sequential levels. First, a binary Convolutional Neural Network (CNN) model classifies the input brain scan image as Normal or Stroke. If stroke-related features are detected, the image is passed to next CNN model for classification into Stage 1, Stage 2, or Stage 3. The system further estimates severity by extracting feature maps from the stage classification CNN and applying K-Means clustering to the extracted deep features. The clustered feature score is combined with stage-based risk priors and prediction confidence to categorize the severity level as Low, Medium, or High. This framework contributes a multi-level stroke screening pipeline by integrating stroke detection, stage classification, and feature-based risk estimation within a single deployed system.