Retrospective Analysis and Data Processing of Machine Learning Techniques for Artefact Detection using Pattern Recognition

A critical phase in the MRI data collection system is the evaluation of quality, involving inspecting the pictures for artefacts and ensuring data quality and the efficacy of any subsequent evaluation or interpretation. High-quality diagnostic images are necessary to enable sophisticated image evaluation techniques for downstream tasks such as segmentation. Among the most important challenges in clinical practice is managing picture artefacts, which can result in poor diagnostic image quality. In this research, we suggest correcting motion-related brain MRI aberrations with a residual U-net design and detecting them with dense convolutional neural networks. We tested our three-class categorization using publicly accessible retrospective datasets, leveraging a 2D MR image. In addition to achieving higher image quality, our pipeline for artefact identification and repair can also achieve higher stroke segmentation accuracy. The method had a 98.34% accuracy rate in identifying artefacts in our studies, and it was validated using a dataset of 29 cases of brain MRI stroke segmentation. We also demonstrated how the suggested correction approach enhanced image quality and segmentation accuracy.