Deep Learning-Based Brain Computed Tomography Image Classification with Hyperparameter Optimization through Transfer...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Interpreting emergent brain CT scans for stroke is time-consuming.
• Deep learning offers efficient computer-aided diagnosis for medical images.

Purpose of the Study:

• To develop and evaluate deep learning models for classifying unenhanced brain CT images.
• To categorize findings into normal, hemorrhage, infarction, and other stroke types.

Main Methods:

• Utilized convolutional neural network (CNN) models: CNN-2, VGG-16, and ResNet-50.
• Employed transfer learning with varied data/mini-batch sizes and optimizers.
• Tested model performance on unenhanced brain CT image classification.

Main Results:

• CNN-2 and ResNet-50 models outperformed VGG-16, achieving an accuracy of 0.9872.
• ResNet-50 had longer processing times compared to other models.
• Developed models demonstrated superior performance over previously reported deep learning methods.

Conclusions:

• Optimized deep learning models can effectively classify brain CT findings for stroke.
• These models can assist clinicians in verifying stroke diagnoses, including hemorrhage and infarction.
• The study highlights the potential of AI in improving the speed and accuracy of stroke diagnosis.