Evaluating Synthetic Medical Images Using Artificial Intelligence with the GAN Algorithm.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Developing synthetic medical images is crucial for research and training.
• Current evaluation methods for synthetic medical images are insufficient, focusing on noise and similarity.
• Existing metrics struggle to detect image duplication and are often time-consuming.

Purpose of the Study:

• To propose a novel method for quantitative and qualitative evaluation of synthetic medical images.
• To address limitations of existing methods, including speed and accuracy in detecting image duplication.
• To provide a reliable tool for assessing the medical suitability of generated images.

Main Methods:

• A hybrid approach combining Feature Matching Distance (FMD) and Convolutional Neural Network (CNN)-based evaluation.
• Comparison of the proposed methods against the Fréchet Inception Distance (FID) method.
• Validation using a dataset of diverse real medical images.

Main Results:

• The FMD method offers significant speed advantages over FID.
• The CNN-based method demonstrates superior accuracy in evaluating synthetic images.
• The combined approach provides a comprehensive assessment of synthetic image quality and reliability.

Conclusions:

• The proposed combined FMD and CNN-based method offers a more effective solution for evaluating synthetic medical images.
• This approach enhances both the speed and accuracy of synthetic image assessment.
• The method reliably evaluates the medical suitability of synthetic images, addressing key limitations in the field.