Deep Learning Models for Detecting Crossbites: Applying Artificial Intelligence  in Orthodontic Diagnostics

Our study published in Head & Face Medicine investigates how deep learning models can assist dentists in diagnosing crossbites using 2D intraoral photographs. We compared six convolutional neural networks to detect and classify non-crossbite, frontal crossbite, and lateral crossbite.

Key Findings

For the Binary Classification (Non-Crossbite vs. Crossbite), the Xception model performed best, correctly identifying crossbites with an accuracy of 98.57% on the test dataset. For the Multiclass Classification (Non-Crossbite vs. Frontal vs. Lateral Crossbite), the DenseNet model achieved the highest accuracy of 91.43%, though the performance dropped slightly when more categories were added.

Implications

This study highlights the high potential of convolutional neural networks (CNNs) for processing clinical photographs and assisting in orthodontic diagnoses. With further refinement, these models could become valuable tools in identifying malocclusions.

Read the full article here.