Evaluating medical images using deep convolutional neural networks: A simulated CT phantom image study.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Deep learning applications

Background:

• Artificial intelligence (AI) research, particularly deep learning, is advancing rapidly.
• AI-driven evaluation of medical images offers potential for significant improvements in examination efficiency.

Purpose of the Study:

• To investigate an AI-based system for evaluating medical image quality.
• Focus on a deep convolutional neural network (CNN) to assess images with varying noise characteristics.

Main Methods:

• Utilized AlexNet (a CNN) trained on natural images and a support vector machine (SVM) for classification.
• Employed synthetic computed tomography (CT) images with controlled signal bodies, contrast, and Gaussian noise for training and testing.

Main Results:

• Two transfer learning approaches were evaluated: SVM classification with AlexNet features and fine-tuning the CNN.
• The SVM method correctly classified all test image noise levels.
• The fine-tuning approach achieved a high accuracy rate of 92.6%.

Conclusions:

• AI-powered image quality evaluation holds promise for clinical applications.
• This technology can be adapted for various image quality indices in the future.