Matched-filter based algorithm for sub- to multi-cellular classification of myocyte tubule-system

In healthy tissue of various mammals, transverse tubules (TTs) are found in a strongly conserved, striated pattern that aligns with the sarcomere network at the z-lines. In a variety of etiologies, this network is perturbed, which is believed to correlate with ineffective, dyssynchronous Ca2+ release and subsequent contraction. Confocal microscopy has become the de facto standard for characterization of TT networks, for which several algorithms for detecting and classifying these subcellular structures have emerged. However, to our knowledge, such algorithms are irrespective of subcellular variations in TT angle and are restricted in application to single isolated myocytes. Here we present a matched filter-based algorithm to characterize TT structure at a subcellular-level, in single cardiomyocytes through millimeter-scale tissue preparations. The algorithm utilizes "filters" representative of intact TT structure, longitudinal remodeling, and TT absence. Application of the algorithm to cardiomyocytes isolated from SHAM, myocardial infarction (MI), and thoracic aortic banding (TAB) animal models confirm and quantify locally-heterogeneous TT structure and structural remodeling. We find significant (p < 0.05) increases in longitudinal remodeling within myocytes intermediate and proximal to an infarct (10 {+/-} 2% and 12 {+/-} 3% compared to 4 {+/-} 2% in SHAM) as well as decreases in tubule striations within 5{degrees} of myocyte minor axis for TAB (36 {+/-} 9%), intermediate (37 {+/-} 4%), and proximal (34 {+/-} 4%) MI myocytes compared to SHAM myocytes (57 {+/-} 12%). Given the reliance of the matched filter approach on images representative of subcellular features, we anticipate the algorithm is generalizable to wide-ranging imaging applications.