Machine learning models in detection and elimination of medical imaging artifacts: Tools in action

Purpose: Machine learning (ML) models have been shown to be effective in eliminating artifacts in medical imaging. Noise is the most common artifact seen, caused by factors such as patient movement, equipment malfunction, and measurement errors. Denoising autoencoders and deep neural network based ML models have shown high efficacy in removing noise from images.

Methods/Materials: Another common artifact is motion blur, which occurs when the patient or the equipment moves during the imaging process. Motion blur detection and correction algorithms have been designed and tested to be efficacious for mammograms and GI imaging.

Results: Metal artifacts are frequently seen in surgical clips/bone and joint implants/dental fillings. Convolutional neural network (CNN) based metal artifact reduction models (CNN-MAR) like beam-hardening correction (BHC) and linear interpolation (LI) have also been in place to reduce implant artifacts from CT images.

Conclusions: Overall, the literature suggests that ML models have the potential to significantly improve the quality of medical images by removing various types of artifacts, thereby increasing diagnostic accuracy and patient outcomes. However, external validation of such models is a need to determine reproducibility of the same.