Analysis of Factors Affecting Quality in Structural Magnetic Resonance Images

Abstract

Combining Magnetic Resonance Images (MRI) from different sources is an increasingly common practice that holds high scientific value. Differences in acquisition parameters and participant characteristics can lead to variations in image quality, highlighting the importance of ensuring these variations do not result in biased statistical outcomes. Here, we investigated contributions of both technical and participant-related factors to MRI quality. We examined how technical factors (scanner hardware, software, and acquisition protocols) affect the Image Quality Rating (IQR) of anatomical MRI. We also evaluated the stability of IQR over time, examined the effects of defacing on image quality, and investigated how participant characteristics (age, sex, and mental health) influence IQR. We collected 2779 T1-weighted volumes, acquired at two different scanner sites (both Siemens 3 Tesla), using two coil array designs (64-channel and 32-channel array), and four scanner software versions (VB17, VB15, VE11, XA30), five acquisition protocols, including two different spatial resolutions (1 mm, 0.8 mm isotropic). Data were collected from 910 healthy controls (HC) (499 women, mean age 27.55 ± 11.27) and from 563 individuals (321 women, mean age 36.42 ± 12.93) with various clinical conditions (125 Major Depressive Disorder [MDD], 43 Autism Spectrum Disorder [AUT], 81 Alcohol Use Disorder [AUD], 104 Schizophrenia [SZ], 70 Chronic Pain [CP], 41 Bipolar Disorder [BD], and 100 with unspecified disease [NHC]). Structural images were preprocessed and analyzed using the quality control pipelines of the Computational Anatomy Toolbox (CAT12, https://neuro-jena.github.io/cat12-help/), which provide an image quality rating (IQR) index for each image, with higher IQR indicating a lower image quality. There was no significant effect of scanner site or coil design on IQR. We found a significant effect of scanner software, with lower image quality for VB17 compared with VB15. There was a significant effect of acquisition protocols (i.e., IQR with protocol “T1_1mm_extended” was higher than with others protocols), and image spatial resolution had a significant impact on IQR, with higher IQR values for 1 mm compared to 0.8 mm. Within participants, IQR was stable across sessions, showing minimal day-to-day variability. Defacing had no significant impact on IQR. Regarding participant characteristics, we observed a significant interaction between sex and age: IQR increased with age in men but not in women. Additionally, participants with SZ had a significant higher IQR compared to HC and MDD. This study provides a comprehensive assessment of the influence of technical and participant-related factors on MRI quality. The findings also support IQR as a robust indicator of image quality and emphasize the importance of integrating image quality metrics, both in multicentric studies and within individual research centers. Incorporating IQR as a quality metric would help minimize biases from image quality variations, enabling a more accurate assessment of underlying structural differences and leading to more reliable findings.