Histogram feature-based classification improves differentiability of early bone healing stages from micro-computed tomographic data
Résumé
Objective: Contrast between not fully mineralized tissues is weak and limits conventional computed tomography (CT). An automated grayscale histogram-based analysis features could improve the sensitivity to tissue alterations during early bone healing. Materials and Methods: Tissue formation in a rat osteotomy model was analyzed using in vivo micro-CT and classified histologically (mineralized, cartilage, and connective tissues). A conventional threshold-based method including manual contouring was compared to a novel moment-based method: after removing the background peak, the histograms of each slice were characterized by their moments and analyzed as a function of the position along the long bone axis. Results: The threshold-based method could differentiate between the mineralized and connective tissue (R2 = 0.73). The moment-based approach yielded a clear distinction between all 3 groups with a classification accuracy up to R2 = 0.93. Conclusions: The moment-based evaluation outperforms the conventional threshold-based CT analysis in sensitivity to the healing stage, user independence, and time consumption.