@article{nokey,

title = {Interpretable machine learning integrates multi-source biomarkers for osteoarthritis diagnosis and mechanistic insights: A temporomandibular joint model},

author = {Najla Al Turkestani and Lucia Cevidanes and Jonas Bianchi and James Sugai and Marcela Gurgel and Juan Prieto and Elizabeth Hatfield and Kristine Philips and Erika Benavides and Fabiana Soki and Yuji Mishina and Margherita Fontana and Arvind Rao and Hongtu Zhu and Tengfei Li},

url = {https://www.sciencedirect.com/science/article/pii/S1063458425011124},

doi = {https://doi.org/10.1016/j.joca.2025.08.002},

issn = {1063-4584},

year  = {2025},

date = {2025-12-01},

journal = {Osteoarthritis and Cartilage},

volume = {33},

issue = {12},

pages = {1522-1533},

abstract = {Objective: Osteoarthritis (OA) is a pan-joint degenerative disorder characterized by cartilage degradation and subchondral bone remodeling. The temporomandibular joint (TMJ) offers a unique model for early OA due to its anatomy and early-onset disease. Current diagnostics rely on late-stage changes, underscoring the need for biomarker integration. We hypothesized that machine learning (ML) combining imaging, molecular, and clinical data would improve diagnostic accuracy, and that SHapley Additive exPlanations (SHAP) would clarify key predictors and interactions, enhancing mechanistic understanding of disease heterogeneity. Design: A case-control study of 162 participants (81 TMJ OA and 81 age- and sex-matched controls) integrated clinical, high-resolution imaging (radiomics, trabecular architecture, joint space), and systemic/articular biomarkers (serum and saliva). Seventy-seven ML combinations were evaluated via nested 10-fold cross-validation. Results: The final ensemble model achieved strong diagnostic performance (AUC=0.828, 95% CI: 0.757–0.892). SHAP analysis revealed top predictors such as headache severity, trabecular thickening, restless sleep, muscle soreness, limited mouth opening and joint space narrowing. Mechanistic interactions captured early inflammatory, structural, and neurovascular changes, including radiomics-cartilage degradation links (e.g., condyle grey level nonuniformity with saliva CXCL-16), clinical-molecular associations (e.g., headaches with saliva VE-cadherin), and subchondral microstructure correlations (e.g., grey level nonuniformity with run length nonuniformity). Conclusions: This study presents a clinically useful, explainable AI model for OA diagnosis. Key predictors and cross-domain interactions improved accuracy and clarified early disease mechanisms. Although cross-validation minimized overfitting risk, external validation is needed. These findings support biomarker-driven precision diagnostics and highlight multi-tissue predictors as potential targets for early OA intervention.},

keywords = {Cartilage degeneration, Diagnostic biomarkers, osteoarthritis, Subchondral bone remodeling},

pubstate = {published},

tppubtype = {article}

}

@article{Bianchi2022e,

title = {Osteoarthritis Diagnosis Integrating Whole Joint Radiomics and Clinical Features for Robust Learning Models using Biological Privileged Information},

author = {Al Turkestani N and Cai L and Cevidanes L and Bianchi J and Zhang W and Gurgel M and Gillot M and Baguero B and Najarian K and Soroushmehr R },

url = {https://www.researchsquare.com/article/rs-1840348/v1},

doi = {https://doi.org/10.21203/rs.3.rs-1840348/v1},

year  = {2022},

date = {2022-07-14},

journal = {Research Square},

abstract = {This paper proposes a machine learning model using privileged information (LUPI) and normalized mutual information feature selection method (NMIFS) to build a robust and accurate framework to diagnose patients with Temporomandibular Joint Osteoarthritis (TMJ OA). To build such a model, we employ clinical, quantitative imaging and additional biological markers as privileged information. We show that clinical features play a leading role in the TMJ OA diagnosis and quantitative imaging features, extracted from cone-beam computerized tomography (CBCT) scans, improve the model performance. As the proposed LUPI model employs biological data in the training phase (which boosted the model performance), this data is unnecessary for the testing stage, indicating the model can be widely used even when only clinical and imaging data are collected. The model was validated using 5-fold stratified cross-validation with hyperparameter tuning to avoid the bias of data splitting. Our method achieved an AUC, specificity and precision of 0.81, 0.79 and 0.77, respectively.},

keywords = {feature selection, machine learning, osteoarthritis, temporomandibular joint},

pubstate = {published},

tppubtype = {article}

}

@article{Bianchi2021b,

title = {Quantitative bone imaging biomarkers to diagnose temporomandibular joint osteoarthritis.},

author = {J Bianchi and J R Goncalves and A C de Oliveira Ruellas and L M Ashman and J-B Vimort and M Yatabe and B Paniagua and P Hernandez and E Benavides and F N Soki and M Loshida and L H S Cevidanes},

url = {https://pubmed.ncbi.nlm.nih.gov/32605824/},

doi = {10.1016/j.ijom.2020.04.018},

year  = {2021},

date = {2021-02-00},

journal = {International Journal of Oral and Maxillofacial Surgery},

volume = {50},

number = {2},

pages = {227-235},

abstract = {Bone degradation of the condylar surface is seen in temporomandibular joint osteoarthritis (TMJ OA); however, the initial changes occur in the subchondral bone. This cross-sectional study was performed to evaluate 23 subchondral bone imaging biomarkers for TMJ OA. The sample consisted of high-resolution cone beam computed tomography scans of 84 subjects, divided into two groups: TMJ OA (45 patients with TMJ OA) and control (39 asymptomatic subjects). Six regions of each mandibular condyle scan were extracted for computation of five bone morphometric and 18 grey-level texture-based variables. The groups were compared using the Mann-Whitney U-test, and the receiver operating characteristics (ROC) curve was determined for each variable that showed a statically significance difference. The results showed statistically significant differences in the subchondral bone microstructure in the lateral and central condylar regions between the control and TMJ OA groups (P< 0.05). The area under the ROC curve (AUC) for these variables was between 0.620 and 0.710. In conclusion, 13 imaging bone biomarkers presented an acceptable diagnostic performance for the diagnosis of TMJ OA, indicating that the texture and geometry of the subchondral bone microarchitecture may be useful for quantitative grading of the disease.},

keywords = {AAOF, Adolescents, biomarkers, Cone-beam computed tomography, Cranial base, osteoarthritis, temporomandibular joint},

pubstate = {published},

tppubtype = {article}

}

@article{Michoud2019,

title = {A web-based system for statistical shape analysis in temporomandibular joint osteoarthritis.},

author = {Michoud L and Huang C and Yatabe M and Bianchi J and et al},

url = {https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6494085/},

doi = {10.1117/12.2506032},

year  = {2019},

date = {2019-05-15},

journal = {Proc SPIE-the Int Soc Opt Eng},

volume = {10953},

abstract = {This study presents a web-system repository: Data Storage for Computation and Integration (DSCI) for Osteoarthritis of the temporomandibular joint (TMJ OA). This environment aims to maintain and allow contributions to the database from multi-clinical centers and compute novel statistics for disease classification. For this purpose, imaging datasets stored in the DSCI consisted of three-dimensional (3D) surface meshes of condyles from CBCT, clinical markers and biological markers in healthy and TMJ OA subjects. A clusterpost package was included in the web platform to be able to execute the jobs in remote computing grids. The DSCI application allowed runs of statistical packages, such as the Multivariate Functional Shape Data Analysis to compute global correlations between covariates and the morphological variability, as well as local p-values in the 3D condylar morphology. In conclusion, the DSCI allows interactive advanced statistical tools for non-statistical experts.},

keywords = {biomarkers, Meshes, osteoarthritis, statistics, temporomandibular joint disorders, web-platform},

pubstate = {published},

tppubtype = {article}

}

@article{Ribera2019,

title = {Shape variation analyzer: a classifier for temporomandibular joint damaged by osteoarthritis},

author = {T R Nina and D Priscille De and Y Marilia and R Antonio and I Marcos and P Beatriz and S Martin and R G Joao and B Jonas and C Lucia and P Juan-Carlos },

url = {https://pubmed.ncbi.nlm.nih.gov/31359900/},

doi = {10.1117/12.2506018},

year  = {2019},

date = {2019-02-00},

urldate = {2019-02-00},

journal = {Dento Maxillo Facial Radiology},

volume = {10950},

abstract = {We developed a deep learning neural network, the Shape Variation Analyzer (SVA), that allows disease staging of bony changes in temporomandibular joint (TMJ) osteoarthritis (OA). The sample was composed of 259 TMJ CBCT scans for the training set and 34 for the testing dataset. The 3D meshes had been previously classified in 6 groups by 2 expert clinicians. We improved the robustness of the training data using data augmentation, SMOTE, to alleviate over-fitting and to balance classes. We combined geometrical features and a shape descriptor, heat kernel signature, to describe every shape. The results were compared to nine different supervised machine learning algorithms. The deep learning neural network was the most accurate for classification of TMJ OA. In conclusion, SVA is a 3D Sheer extension that classifies pathology of the temporomandibular joint osteoarthritis cases based on 3D morphology.},

keywords = {Classification, Deep Learning, Neural Network, osteoarthritis, temporomandibular joint disorders},

pubstate = {published},

tppubtype = {article}

}