MRI-derived neuroimaging biomarkers —such as volumes of brain ssue, white maer lesions, or tumors— play a major role in the early diagnosis and staging of disease. Image quality is a key determinant of the sensivity of these biomarkers. To improve image quality, we have to bring advanced image QC methods to clinical research, pracce, and trials, which this project achieves by addressing the following key challenges of current QC methods:
● Visual rang — the current gold standard — has large intra- and inter-observer variability;
● Image quality is not fully quanfied but based on the presence of clear arfacts;
● Quality thresholds for approving or rejecng image data are not established.
Our work focus on:
We will develop a QC strategy based on large mul-scanner and mul-cohort clinical datasets (>10,000 images) that we enrich with varying degrees of simulated MRI arfacts4,5, for having both real-world data and ground truth. Addionally, we will combine previously developed image QC features from our commercial partners5–7, ourselves5,8, and those publicly available9–13 to quanfy image quality.
Image quality thresholds are a trade-off between sample size and image quality, which we will opmize based on established physiological associaons. We will study the bias and variability in predicng age and other demographics14,15 in images from healthy volunteers enriched by our simulated arfacts.
We will assess the ability of our QC strategy to handle simulated arfacts in structural MRI images and differenate them from pathology as evaluated through clinical associaons with gray maer atrophy in Alzheimer’s Disease, lesion volume in MS, and brain tumor volume.