Lightweight deep learning for medical imaging using MobileNetV2-based brain pathology classification with Grad-CAM...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Computed Tomography (CT) brain scans are vital for diagnosing neurological conditions like tumors and aneurysms.
• Accurate classification of CT brain scans aids in treatment decisions and disease monitoring.
• Enhancing the accuracy and interpretability of brain abnormality detection is a key clinical challenge.

Purpose of the Study:

• To develop and evaluate a novel framework for brain CT image classification.
• To improve the accuracy and interpretability of detecting brain abnormalities using deep learning.
• To compare the performance of multiple pre-trained deep learning models for this task.

Main Methods:

• Utilized a comprehensive dataset of CT brain scans with Digital Imaging and Communications in Medicine (DICOM) pre-processing.
• Employed transfer learning with four state-of-the-art pre-trained models: MobileNetV2, ResNet-50, EfficientNet-B0, and VGG-16.
• Incorporated Grad-CAM for model interpretability and ensemble prediction techniques.

Main Results:

• MobileNetV2 and the Ensemble model achieved the highest classification accuracy (97.44%) and macro-AUC scores (0.9895-0.9914).
• VGG16 showed strong performance with 92.31% accuracy and the highest macro-AUC (0.9962).
• MobileNetV2 demonstrated exceptional efficiency with rapid training (89 s) and testing (38.16 s) times.

Conclusions:

• MobileNetV2 offers a highly accurate and computationally efficient solution for brain CT image classification.
• The proposed framework enhances the detection of brain abnormalities, with potential for clinical deployment.
• The study highlights the effectiveness of transfer learning and ensemble methods in medical image analysis.