No-Reference Image Quality Assessment Of Large FLAIR MRI Datasets

The study of neurodegenerative diseases have found promise through white matter lesions best visualized in FLAIR MRI; however, algorithms experience difficulty in generalizing to large multicenter datasets due to the variance of image quality and characteristics. This thesis presents a quality control tool that combines image quality assessment with outlier rejection algorithms; this tool is unique as it is specifically designed for large multicenter FLAIR MRI datasets. An image processing approach evaluates each volume by: intensity-based features, sharpness/blur-based features, signal- and contrast-to-noise ratios, noise field characteristics, motion artifact prevalence

and a total IQ score. The performance of this tool was evaluated on labelled ADNI and CCNA data reporting F1 scores of 0.82, and 0.85, respectively. Applications for this tool include potential rescan or longitudinal scanner study alongside the immediate application of outlier removal for

large FLAIR datasets.