Interpretable Deep Neural Networks and Bayesian Inference for Orthodontics

EasyChair Preprint 15221

Abstract

Artificial Intelligence (AI), Machine Learning (ML), and more specifically Deep Neural Network (DNN) have enabled numerous applications in health related science, and more specifically in dentofacial orthopedics, for diagnosis and clinical decision-making support. DNNs and particularly Convolutional Neural Networks (CNNs), powered by the perturbation and saliency maps such as Class Activation Mapping (CAM) and its variants such as Score-CAM have become very popular tools, in medical community thanks to the interpretability of these methods.

They are used in particular to locate biomarkers, i.e., observable or measurable signals that reflect the presence, severity or evolution of a disease and that can be considered as discriminating regions in abnormal images.

For orthodontics domain, recent works proposed an innovative methodology coupling DNN and interpretability algorithms to assess cranofacial structures impacted by mandibular retrognathia. Applied to a set of radiographs classified into physiological versus pathological categories, this methodology made it possible to discuss the structures impacted by retrognathia, and to identify new structures of potential interest in medical terms, and to highlight the dynamic evolution of impacted structures according to the level of gravity of mandibular retrognathia.

The main object of this paper is to show how the Interpretable Deep Learning methods combined with Bayesian inference, can be a real "Third Eye" for medical and biological diagnostics in general, and in particular for the orthodontics diagnosis.

Keyphrases: Bayesian inference, Cephalometry, Interpretable Deep Learning Neural Network Artificial Intelligence, Orthodontics diagnosis

@booklet{EasyChair:15221,
  author    = {Masrour Makaremi and Alireza Vafaei Sadr and Ali Mohammad-Djafari},
  title     = {Interpretable Deep Neural Networks and Bayesian Inference for Orthodontics},
  howpublished = {EasyChair Preprint 15221},
  year      = {EasyChair, 2024}}