Ischemic stroke lesion detection, characterization and classification in CT images with optimal features selection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Ischemic stroke is a leading cause of death and disability.
• Computed tomography (CT) is crucial for rapid diagnosis and treatment planning.
• Accurate segmentation and classification of ischemic lesions are vital.

Purpose of the Study:

• To develop a novel histogram bin-based algorithm for segmenting ischemic stroke lesions in CT images.
• To identify optimal feature groups for classifying normal and abnormal brain regions.
• To improve the accuracy of ischemic stroke detection using machine learning classifiers.

Main Methods:

• Image pre-processing and segmentation of ischemic lesions.
• Extraction of texture features including first-order, gray level run length matrix, gray level co-occurrence matrix, and Hu's moments.
• Feature ranking, grouping, and selection for optimal classification.
• Classification using logistic regression, support vector machine, random forest, and neural network classifiers.

Main Results:

• The proposed algorithm effectively segments and classifies ischemic stroke lesions.
• Optimal feature group FG12 achieved high classification accuracies: 88.77% (LR), 97.86% (SVMC), 99.79% (RFC), and 99.79% (NNC).
• Results were validated using fourfold cross-validation, demonstrating robustness.

Conclusions:

• The developed histogram bin-based algorithm with optimal feature group selection is effective for ischemic stroke detection.
• Machine learning classifiers, particularly Random Forest and Neural Network, show high performance in classifying stroke lesions.
• This approach offers a promising tool for enhancing the diagnostic accuracy of ischemic stroke from CT images.