Machine Learning-Assisted SHG Morphometry Reveals Distinct Collagen Microarchitectures of Trabecular Bone and Fibrosis in Bone Marrow Biopsies.

Collagen microarchitecture in bone marrow biopsies represents a largely underexplored source of candidate quantitative biomarkers for histopathological diagnostics and analysis of tissue remodeling. While second harmonic generation (SHG) microscopy has been increasingly applied to fibrosis assessment, the collagen organization of trabecular bone in bone marrow trephine biopsies remains poorly characterized. Here, we combined high-resolution SHG microscopy with shallow machine learning-assisted morphometry to compare collagen architecture in structured trabecular bone, unstructured trabecular bone, and fibrosis in bone marrow biopsies from patients with primary myelofibrosis. SHG image segmentation was performed using the LabKit plugin in Fiji. Several annotation strategies were evaluated to identify classifier configurations that preserved fibrillar structures. Quantitative morphometric analysis revealed marked differences in collagen organization between tissue types. Per-patient analysis consistently demonstrated thinner collagen fibers and reduced branching complexity in fibrosis than in structured trabecular bone. In contrast, unstructured trabecular bone showed extensive network branching accompanied by shorter skeleton branch length, consistent with remodeling-associated alterations of trabecular collagen architecture. Our results further demonstrate that annotation strategy substantially influences segmentation outcome and downstream morphometric measurements in SHG-based collagen analysis. Overall, this descriptive proof-of-concept study establishes a reproducible workflow for machine learning-assisted SHG morphometry that may prove useful for quantitative assessment of fibrosis, bone remodeling, and extracellular matrix organization in bone marrow pathology.