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

Abstract

Objective: To develop and validate a machine learning (ML) approach for automatic three-dimensional (3D) histopathological grading of osteochondral samples imaged with contrast-enhanced microcomputed tomography (CEmCT). Design: A total of 79 osteochondral cores from 24 total knee arthroplasty patients and two asymptomatic donors were imaged using CEmCT 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 CEmCT grades (diameter ¼ 2 mm samples). Models were validated using nested leave-one-out crossvalidation 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 ¼ 0.49, 0.69 and 0.71 for SZ, DZ and CZ, respectively) and LR (AP ¼ 0.9 [0.77e0.99], 0.46 [0.28e0.67] and 0.65 [0.41e0.85] 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 ¼ 0.85 [0.73e0.93], 0.82 [0.70e0.92] and 0.8 [0.67e0.9], 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.

Predicting total knee arthroplasty from ultrasonography using machine learning

“…regularization coefficient in logistic regression), and assess the model performance without overfitting. Here, we applied the nested leave-one-out cross-validation procedure that allowed to overcome the biases of model assessment in the small-data regime [ 37 ]. Furthermore, we used a simple linear model – logistic regression, to combat overfitting.…”

Automated Prediction of Osteoarthritis Level in Human Osteochondral Tissue Using Histopathological Images

“…However, few studies have looked at automation in the grading of histopathological samples. Rytky et al used regularized linear and logistic regression models for the histopathological grading of osteochondral specimens imaged with contrast-enhanced microcomputed tomography (microCT) [23]. The models were trained against the manually graded histopathological samples to predict the grades of degeneration for the articular cartilage of the surface, deep, and calcified cartilage zone.…”

“…The models were trained against the manually graded histopathological samples to predict the grades of degeneration for the articular cartilage of the surface, deep, and calcified cartilage zone. They found that the model could detect the degeneration in the surface zone with an average precision of 0.89 (AUC of 0.92) while the detection of degeneration in the deep zone was the lowest, with an average precision of 0.46 (AUC of 0.62) [23]. Power et al used supervised deep learning to automate the grading system for the histological images of engineering cartilage tissue [24].…”

Automated analysis of rabbit knee calcified cartilage morphology using micro‐computed tomography and deep learning

“…We showed previously that µCT images of the human subchondral plate contain both the mineralized CC and the subchondral bone (Finnilä et al, 2017 ). Indeed, CC cannot be separated from bone with low‐resolution µCT imaging but becomes visible only in high‐resolution µCT images (Rytky et al, 2020 ). However, because of the very minor difference in mineralization between the subchondral bone and CC, it is challenging to delineate the interface between CC and subchondral bone also in high‐resolution µCT imaging.…”