|
Abstract:
|
In a medical imaging dataset, there are many ways in which statistical issues may arise. Data monitoring and quality control (QC) play a large role in the accuracy of big data. Outliers in the dataset must be checked, as they could be a systemic error or the natural history of the disease. We are trying to identify the simplest method to determine outliers for QC of disease growth across multiple timepoints. We analyzed three individual cohorts with different disease types: volumetric tumor growth for low-grade and high-grade glioma, and fibrotic disease growth of the lung for idiopathic pulmonary fibrosis. We modified the longitudinal data for subjects who had irregular observation [Pullenayegum and Lim (2016)]. Using a mixed effect model on percent change of disease growth, we can identify outliers based on the random effects by using Cook’s distance. We could potentially provide automated QC to investigate the outliers during the radiographic read to determine legitimacy of the disease or to fix the possible measurement error. This approach could apply as a general QC across a large variety of studies with disease growth quantified from longitudinal radiographic imaging data.
|
