Automated assessment of image quality metrics in brain magnetic resonance imaging - MNUMS’s Mongol-Japan Hospital

Abstract

Background: Magnetic resonance imaging (MRI) is a powerful diagnostic tool used in central nervous system disorders. Quality assurance is required during the scan and the interpretation processes. Due to long scan times, there is a possibility of image artefacts resulting from patient movements and the MRI machine. It is the responsibility of the radiologist and technician to distinguish between them. However, with large-scale studies involving images from different sites and machines, the process of image quality assessment takes even longer, thus an automated assessment of image quality is essential for quantitative neuroimaging research. Aim: We aimed to compare the image quality metrics of brain MRI images with radiologist’s assessment and to determine effect of some patient-related factors on image quality Materials and Methods: This retrospective descriptive study included MRI scans performed at the MNUMS’s Mongolia-Japan Hospital by Canon Vantage-Elan 1.5T. The image quality metrics were calculated using the MRIQC software. The following image quality metrics were used in this study: • CJVcoefÏcient of joint variation between GW and WM • CNRcontrast to noise ratio • ICVIntracranial volume • SNRsignal to noise ratio Results: This study enrolled 200 participants, with female: male ration of 69:31, with average age of 43.38±12.54. Participants were divided into 4 age groups: 18-30(16%), 31-45(38%), 46-55(28%), and 56-65(18%). There was no significant difference between men and women in terms of intracerebral volume p>0.901. To determine inter-rater reliability of the 2 radiologists participating, we used Cohen’s kappa for each of the seven 7 criteria (Motion artefact k-value – 0.173, Signal to noise ratio k-value -0.155, Image contrast k-value – 0.111, Aliasing/ Ghosting k value – 0.01, non-uniform intensity k-value – 0.125, Zipper artefact k-value – 0.128, Metal artefact k-value - 0.448). To assess the effect of comfort on image quality, we had 100 of the participants complete a 4-point Likert scale questionnaire with 7 questions after the scan. Afterwards we divided the patients into two groups depending on the point with Group A -1-14 point (68%) being the most comfortable and Group B – 15-28 point (32%) being less comfortable during the scan. There was no statistically siginificant correlation between patient comfort and image quality metrics. Conclusion: All seven image quality assessment parameters were significant when comparing the inter-rater reliability of the participating radiologists, with metal artifact having the highest value (k value-0.448, p<0.05.) Aging demonstrated an increase in CSF volume and decrease of gray matter volume. Patients’ perceived comfort level as well as the better knowledge about MRI procedure execution tend to have positive effect on image signal-to-noise ratio and tissue contrast.