Automated Segmentation of Tissue Zones in Distraction Osteogenesis

Automated analysis of histological sections in distraction osteogenesis can reduce manual effort and subjectivity in histological assessment. A 2D nnU-Net was trained to segment three characteristic tissue zones in sections of rat tibial distraction osteogenesis using hematoxylin and eosin (H&E) staining: the fibrous interzone (FIZ), the microcolumn formation zone (MCF), and the remodeling zone (RM). Model performance was evaluated using five-fold cross-validation and Dice loss. The model achieved reliable performance for FIZ and MCF, while RM remained the most challenging region due to lower frequency, weak boundaries and morphological variability. The results establish a proof of principle for applying deep learning to histological zone segmentation, supporting more objective and scalable analysis.