Inter-reader variability for quality assessment of synthetically degraded prostate MR images aligned with PIQUAL...

The purpose of this study was to evaluate expert radiologists' perceptions of the quality of magnetic resonance imaging (MRI) of the prostate for images that underwent controlled degradation.

Seven-levels of controlled degradations were applied to 10 single-slice image sets, with T2-weighted images degraded using Gaussian noise and blur, and diffusion-weighted images degraded using noise, blur, and elastic deformation. Twelve image sets of varying apparent-diffusion coefficient (ADC)-T2-weighted anatomical mismatch were also included. Ten expert radiologists independently assessed image quality using prostate imaging quality (PI-QUAL) aligned criteria. Inter-reader agreement was evaluated using proportion agreement and Gwet's agreement coefficient 1 (AC1).

For T2-weighted images, image quality adequacy declined with increasing noise and blur, with inter-reader agreement highest at minimal and severe degradation (AC1 up to 0.98), and lowest at intermediate levels (AC1 as low as -0.03), demonstrating a U-shaped pattern. For diffusion-weighted imaging, image quality adequacy similarly decreased with increasing blur, and warping, while agreement for noise-degraded images increased from moderate at baseline to substantial at the highest noise levels. ADC- T2-weighted alignment agreement was almost perfect for mismatches < 5 mm (AC1, 0.93), but poor for > 5 mm mismatches (AC1, -0.08).

Expert radiologists' assessment of image quality is subjective, with inter-reader agreement varying with image quality and type. Higher agreement is demonstrated for extreme MRI examinations of optimal or markedly poor quality, with the greatest disagreement at moderate degradation levels. Diffusion-weighted images, in particular, show more variable assessments, highlighting its susceptibility to noise and artifacts.