A web-based system for neural network based classification in temporomandibular joint osteoarthritis

Area of Science:

• Biomedical Data Science
• Medical Imaging Analysis
• Osteoarthritis Research

Background:

• Temporomandibular joint osteoarthritis (TMJOA) diagnosis relies on clinical, imaging, and biological markers.
• Integrating diverse data types for TMJOA analysis presents significant computational challenges.

Purpose of the Study:

• To introduce methodological innovations in a web-based system for biomedical data storage, integration, and computation.
• To utilize this system for training a deep neural network classifier for TMJOA using imaging data.

Main Methods:

• Development of a web-based system for data storage, computation, and integration (DSCI).
• Implementation of a deep neural network classifier (Shape Variation Analyzer, SVA) for 3D condylar morphology.
• Utilized cone beam computed tomography (CBCT) scans to create 3D surface meshes of mandibular condyles.

Main Results:

• The DSCI system successfully trained and tested a neural network, classifying 5 stages of TMJ structural degenerative changes with 91% agreement with clinician consensus.
• A novel statistical analysis (Multivariate Functional Shape Data Analysis) was applied via DSCI to compute high-dimensional correlations between 3D shape, pain levels, and biological markers.

Conclusions:

• The study demonstrates a comprehensive phenotypic characterization of TMJ health and disease.
• Novel, flexible, open-source tools were developed for a web-based system enabling advanced statistical analysis and neural network classification of TMJOA.