Detection of Subarachnoid Hemorrhage Using CNN with Dynamic Factor and Wandering Strategy-Based Feature Selection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Subarachnoid Hemorrhage (SAH) presents a high mortality rate, necessitating rapid and accurate detection.
• Challenges in SAH detection include noisy CT images, limited data, and model training issues like overfitting and vanishing gradients.

Purpose of the Study:

• To develop an automated system for expedited and improved detection and grading of SAH.
• To address limitations of existing methods by enhancing feature selection and classification accuracy.

Main Methods:

• A novel DWSCSO (water waves dynamic factor and wandering strategy-based Sand Cat Swarm Optimization) was used for optimal feature selection.
• A PRSCNN (Parametric Rectified Linear Unit with Stacked Convolutional Neural Network) was developed for SAH grading.
• Region growing segmentation and feature extraction from pre-trained models (GoogleNet, VGG-16, ResNet50) were employed.

Main Results:

• The DWSCSO-PRSCNN model achieved a maximum accuracy of 99.48% in SAH detection.
• The model demonstrated improved accuracy of 99.62% on a CT dataset compared to existing methods.
• The approach effectively reduced false positives and false negatives in SAH classification.

Conclusions:

• The DWSCSO-PRSCNN model offers a robust solution for SAH detection, overcoming challenges like overfitting and vanishing gradients.
• The developed method enhances feature selection and classification, proving essential for clinical SAH detection.
• The complexity of the DWSCSO-PRSCNN is deemed acceptable, providing superior performance over simpler methods.