Image Quality Assessment for Magnetic Resonance Imaging

Abstract

Image quality assessment (IQA) algorithms aim to reproduce the human's perception of the image quality. The growing popularity of image enhancement, generation, and recovery models instigated the development of many methods to assess their performance. However, most IQA solutions are designed to predict image quality in the general domain, with the applicability to specific areas, such as medical imaging, remaining questionable. Moreover, the selection of these IQA metrics for a specific task typically involves intentionally induced distortions, such as manually added noise or artificial blurring; yet, the chosen metrics are then used to judge the output of real-life computer vision models. In this work, we aspire to fill these gaps by carrying out the most extensive IQA evaluation study for Magnetic Resonance Imaging (MRI) to date (14,700 subjective scores). We use outputs of neural network models trained to solve problems relevant to MRI, including image reconstruction in the scan acceleration, motion correction, and denoising. Our emphasis is on reflecting the radiologist's perception of the reconstructed images, gauging the most diagnostically influential criteria for the quality of MRI scans: signal-to-noise ratio, contrast-to-noise ratio, and the presence of artifacts. Seven trained radiologists assess these distorted images, with their verdicts then correlated with 35 different image quality metrics (full-reference, no-reference, and distribution-based metrics considered). The top performers -DISTS, HaarPSI, VSI, and FID VGG16 -are found to be efficient across three proposed quality criteria, for all considered anatomies and the target tasks.

Pytorch Image Quality: Metrics for Image Quality Assessment

“…Note that, in our results, the SRCC score was reported as a measure of agreement with the user opinion (MOS); we consider this score the most relevant for our study. We find Pearson's linear correlation coefficient to be less descriptive because MOS and the metrics' values are typically non-linearly correlated (refer to [15], also confirmed herein empirically). Conversely, KRCC is highly-correlated with SRCC, as they both correspond to the same (rank order) family of methods.…”

Synthesizing fractional anisotropy maps from T1-weighted magnetic resonance images using a simplified generative adversarial network

“…They are prone to misjudging brightness shifts or blurring, erroneously categorizing low-quality images as high quality. 24 Future work should focus on exploring alternative datasets to enhance generalizability and robustness of the proposed model. Furthermore, it would be beneficial to explore alternative performance metrics specifically tailored for MRI medical images.…”

ProactiV: Studying Deep Learning Model Behavior Under Input Transformations

“…Transforms that do not have continuity, like adversarial perturbations, are out of scope. Users can also define custom difference metrics [55] to study desired behavior. Advanced concept analysis [21], [22] can support intermediate group-level analysis.…”