Deep learning-driven 3-D histopathology method: a pipeline for cellular-resolution myocarditis analysis

Infiltrative cardiovascular diseases are characterized by complex spatial distributions of abnormal tissue that hinder accurate diagnosis and therapeutic monitoring. Conventional two-dimensional (2-D) histopathological analysis often fails to capture their full extent, limiting quantitative assessment of disease burden and treatment response. We present a Linux-optimized, Cellular-resolution Organ Digital Analysis (CODA) pipeline for quantitative three-dimensional (3-D) reconstruction of whole murine hearts at cellular resolution. Our adaptations ensure compatibility with Linux environments, incorporate visualization tools for integration with external software, and integrate myocarditis-specific training data to improve segmentation accuracy, along with the standard triangle language (STL) format export for visualization in 3-D Slicer. Applied to a COVID-19-induced myocarditis model (n = 6), the modified pipeline achieved >90% accuracy in segmenting affected tissue, revealing significantly greater volumes of inflammatory and necrotic foci compared with controls. This open-source platform provides a scalable, pathologist-artificial intelligence (AI) hybrid workflow for precise 3-D tissue analysis across infiltrative cardiac diseases. NEW & NOTEWORTHY We introduce a Linux-compatible, myocarditis-specific CODA pipeline, integrating DeepLabv3+ segmentation, parallel processing (<150 min per heart), and STL-based 3-D visualization. With over 90% accuracy in identifying affected tissue, it uncovered increased inflammatory and necrotic regions in a viral myocarditis model versus controls. This open-source, deep learning-powered tool bridges computational efficiency and translational relevance for quantitative 3-D analysis of infiltrative heart disease.