A Mixture Modeling Approach to Image Normalization in Highly Multiplexed Cellular Imaging Data

• Activity Number: 47 - Recent Development in Imaging Statistics
• Type: Contributed
• Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Coleman Harris* and Eliot McKinley and Joseph Roland and Ken Lau and Robert Coffey and Simon Vandekar
• Companies: Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University Medical Center and Vanderbilt University
• Keywords: imaging; normalization; multiplexed; mixture model; batch effects; intensity

Abstract

There is interest in using high-dimensional multiplexed imaging methods to quantify the heterogeneity of cell populations in healthy & tumor tissue to understand tumor progression & improve treatment strategies. However, implicit biases exist in the imaging pipeline – images are distorted by optical effects, slide/batch effects, & instrument variability. The challenge in normalizing multiplexed imaging is compounded by the number of channels & natural tissue variability within each field of view, which can bias downstream analyses by introducing systematic differences in image intensity that can impact inference. Here we introduce an image normalization pipeline to reduce systematic variability in multiplexed image intensity. We utilize existing neuroimaging methods & develop a Gaussian mixture model normalization procedure to reduce the systematic variability of the image intensity across slides. We demonstrate these methods by analyzing cells segmented in multiplexed immunofluorescence (MxIF) images to study the reduction in variability. This pipeline can be implemented to reduce these systematic errors & offer insight into ways the image analysis pipeline can be improved.