HOME   |    PDF   |   


Title

Artificial intelligence-based prediction of orthodontic bracket bond failure using deep learning: An in vitro study

 

Authors

Manish Suresh Agrawal 1, Pranav R. Kamble 2, Prerna Raje Batham 3, Ashok Panika 3, Akshay Gupta 4 & Shobhit Saxena 5,*

 

Affiliation

1Department of Orthodontics and Dentofacial Orthopaedics, Bharati Vidyapeeth (Deemed to be) University Dental College and Hospital, Sangli, Maharashtra, India; 2Department of Orthodontics and Dentofacial Orthopedics, Nanded Rural Dental College, Nanded, Maharashtra, India; 3Department of Orthodontics and Dentofacial Orthopaedics, Government College of Dentistry, Indore, Madhya Pradesh, India; 4Department of Orthodontics and Dentofacial Orthopedics, Indira Gandhi Government Dental College, Amphalla, Jammu 180001, India; 5Department of Orthodontics and Dentofacial Orthopedics, Narsinhbhai Patel Dental College and Hospital, Sankalchand Patel University, Visnagar, Gujarat, India; *Corresponding author

 

Email

Manish Suresh Agrawal - E-mail: drmanishortho2011@gmail.com
Pranav R. Kamble - E-mail: pranavkamble679@gmail.com
Prerna Raje Batham - E-mail: drprernaraje@gmail.com
Ashok Panika - E-mail: drashokpanika@gmail.com
Akshay Gupta - E-mail: drakshay.ortho@gmail.com
Shobhit Saxena - E-mail: shobhit.fds@spu.ac.in
 

Article Type

Research Article

 

Date

Received October 1, 2026; Revised October 31, 2026; Accepted October 31, 2026, Published October 31, 2026

 

Abstract

Bracket bond failure can prolong orthodontic treatment and reflects interacting effects of enamel condition, bonding variables, aging and interface defects. Therefore, it is of interest to develop the internally validated a multimodal deep-learning model using standardized enamel-bracket interface images and laboratory variables from 240 bonded premolars. On the held-out test set, the multimodal model achieved an area under the ROC curve of 0.92, accuracy of 86.1%, sensitivity of 88.2% and specificity of 84.2%, outperforming the image-only model and logistic regression comparator. Saliency analysis most frequently highlighted interface discontinuities and marginal voids, while addition of experimental variables increased AUC by 0.06. Thus, the model showed promising laboratory discrimination and calibration, but external and prospective validation across different adhesives, brackets, operators and imaging systems is required before any clinical application.

 

Keywords

Artificial intelligence (AI), deep learning (DL), orthodontic bracket, bond failure, shear bond strength, thermocycling

 

Citation

Agrawal et al. Bioinformation 22(10): 7493-7497 (2026)

 

Edited by

Ashwini Dhopte

 

ISSN

0973-2063

 

Publisher

Biomedical Informatics

 

License

This is an Open Access article which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. This is distributed under the terms of the Creative Commons Attribution License.