Automating three-dimensional osteoarthritis histopathological grading of human osteochondral tissue using machine learning on contrast-enhanced micro-computed tomography

Objective: To develop and validate a machine learning (ML) approach for automatic three-dimensional (3D) histopathological grading of osteochondral samples imaged with contrast-enhanced micro-computed tomography (CEμCT). Design: A total of 79 osteochondral cores from 24 total knee arthroplasty patients and two asymptomatic donors were imaged using CEμCT with phosphotungstic acid -staining. Volumes-of-interest (VOI) in surface (SZ), deep (DZ) and calcified (CZ) zones were extracted depth-wise and subjected to dimensionally reduced Local Binary Pattern -textural feature analysis. Regularized linear and logistic regression (LR) models were trained zone-wise against the manually assessed semi-quantitative histopathological CEμCT grades (diameter = 2 mm samples). Models were validated using nested leave-one-out cross-validation and an independent test set (4 mm samples). The performance was primarily assessed using Mean Squared Error (MSE) and Average Precision (AP, confidence intervals are given in square brackets). Results: Highest performance on cross-validation was observed for SZ, both on linear regression (MSE = 0.49, 0.69 and 0.71 for SZ, DZ and CZ, respectively) and LR (AP = 0.9 [0.77–0.99], 0.46 [0.28–0.67] and 0.65 [0.41–0.85] for SZ, DZ and CZ, respectively). The test set evaluations yielded increased MSE on all zones. For LR, the performance was also best for the SZ (AP = 0.85 [0.73–0.93], 0.82 [0.70–0.92] and 0.8 [0.67–0.9], for SZ, DZ and CZ, respectively). Conclusion: We present the first ML-based automatic 3D histopathological osteoarthritis (OA) grading method which also adequately perform on grading unseen data, especially in SZ. After further development, the method could potentially be applied by OA researchers since the grading software and all source codes are publicly available.