A Comparative Analysis of Techniques in Structural Brain MRI Quality Assessment

The quality of Structural Magnetic Resonance Imaging (sMRI) scans plays a crucial role in ensuring the reliability of diagnostic and research outcomes. Low-quality images can introduce bias, affecting downstream analyses and clinical decision-making. This review presents an overview of the various techniques available for the automatic quality assessment (QA) of the Brain MRI scans developed so far. The aim of this review was to highlight the differences between these techniques based on their key features, performance metrics and evaluation benchmarks.A systematic search was performed across academic databases, including PubMed, Web of Science, IEEE Xplore and Google Scholar. Additionally, reference lists from the retrieved articles were manually screened to identify relevant studies using inclusion and exclusion criteria.A total of 14 different techniques were identified, each exhibiting notable differences in methodologies, features, datasets and benchmarks. These techniques were grouped into three categories: conventional approaches, classical machine learning methods and deep learning techniques. Quality metrics were used to assess the quality of the scan. Lack of standardized datasets and evaluation metrics hinders the direct comparison of these methods. But an attempt was made to compare these methods based on accuracy and number of scans used for training.The review underscores the strengths and limitations of QA techniques, highlighting the critical need for standardization in datasets and evaluation metrics. Advancing QA in sMRI requires developing measures independent of dataset and algorithm to enhance comparability and reliability in clinical and research contexts.