Blind CT Image Quality Assessment Using DDPM-Derived Content and Transformer-Based Evaluator.

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Low-dose CT scans reduce radiation exposure but often produce images with noise and artifacts.
• Blind Image Quality Assessment (BIQA) is crucial for evaluating and improving low-dose CT reconstruction techniques.
• Existing BIQA methods can be enhanced by mimicking the human visual system's (HVS) perceptual processing.

Purpose of the Study:

• To develop an innovative BIQA metric that emulates the internal generative mechanism (IGM) theory of the HVS.
• To improve the accuracy of perceptual quality assessment for low-dose CT images.

Main Methods:

• An active inference module using a denoising diffusion probabilistic model (DDPM) was developed to predict primary image content.
• A dissimilarity map was generated by comparing the distorted image with its predicted primary content.
• A transformer-based image quality evaluator processed a multi-channel image combining the distorted image and dissimilarity map.

Main Results:

• The proposed BIQA metric demonstrated competitive performance on a low-dose CT dataset.
• Leveraging DDPM-derived primary content improved the assessment of image quality.

Conclusions:

• The developed BIQA metric effectively emulates human visual perception for low-dose CT images.
• This approach offers a promising direction for advancing low-dose CT image reconstruction and quality assessment.