Theia: Bleed-Through Estimation With Interaction Terms And Convolutional Kernels

Advances in multiplexed fluorescent microscopy have significantly increased the quality and utility of multi-channel biomedical images. With these advances, the problem of bleed-through across channels has become increasingly visible. Bleed-through negatively impacts analyses such as fluorescent marker correlation and cellular and subcellular segmentation. We present Theia, a LASSO based model for bleed-through estimation. Theia intelligently selects regions of interest in images, uses interaction terms to account for objects tagged with multiple fluorophores, accounts for pixel-scale spatial uncertainties with convolutional kernels, and is highly performant even on inexpensive computing hardware. Theia also exhibits excellent performance as measured by Pearson Correlation.