Intracranial hemorrhage detection in 3D computed tomography images using a bi-directional long short-term memory...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Accurate detection of intracranial hemorrhage (ICH) in 3D Computed Tomography (CT) brain images is crucial but challenged by limited labeled data.
• Existing machine learning models often struggle with the complexity and variability of ICH presentation in medical imaging.

Purpose of the Study:

• To develop and evaluate a novel model for enhanced intracranial hemorrhage detection in 3D CT brain images.
• To address the challenge of scarce labeled data in medical imaging for ICH classification.

Main Methods:

• A new model was implemented, utilizing Otsu's thresholding for region of interest (RoI) segmentation.
• Feature extraction was performed using Tamura features and Gradient Local Ternary Pattern (GLTP) descriptors.
• A modified genetic algorithm with infinite feature selection reduced feature redundancy, followed by classification using a Bi-directional Long Short Term Memory (Bi-LSTM) network.

Main Results:

• The Bi-LSTM based modified genetic algorithm achieved superior performance compared to traditional machine learning models.
• The model demonstrated high sensitivity (99.40%), accuracy (99.80%), and specificity (99.48%) in detecting and classifying ICH subtypes.

Conclusions:

• The proposed Bi-LSTM model integrated with a modified genetic algorithm offers a highly effective solution for intracranial hemorrhage detection.
• This approach significantly overcomes data scarcity issues and outperforms existing methods in classifying various ICH subtypes.