Explainable AI for predicting pulp exposure during caries treatment

Overview

This research artifact, developed by the University of Copenhagen, is an artificial intelligence-based solution designed to predict the probability of pulp exposure before caries excavation in teeth with extensive decay. By analyzing preoperative bitewing radiographs, the system assists dental professionals in making more informed decisions regarding treatment modalities—such as stepwise versus non-selective excavation—thereby aiming to reduce unnecessary invasive procedures.

Technical specifications

The AI framework utilizes a multi-path neural network architecture:

• Convolutional Neural Network (CNN): Based on the ResNet-50 architecture to analyze radiographic image data.
• Segmentation Analysis: A second network path processes the distance between the pulp and the lesion extracted from X-ray segmentations.
• Data Integration: The two paths merge and are processed through fully connected layers to output a probability prediction of pulp exposure.
• Input data includes radiographic images, patient age, and reported pain levels.

Technology readiness level

The technology is at level 4. It has been developed and validated through laboratory-based studies and proof-of-concept randomized controlled trials involving dental students. While the model has demonstrated better performance than students in experimental settings, researchers note that further development is required to improve explainability and to fully address the challenges of clinical adoption and real-world practice integration.