Towards an Automated Caries Detection System Using Intra-Oral Photographs

Objective: ,
The problem of detecting caries as a major challenge in the field was reviewed by a panel of international experts in 2001. The resulting standard, published in 2004, named the International Caries Detection and Assessment System (ICDAS), provides guidelines for detection of caries. This study aims to develop an automated image-processing/machine-learning method that processes intra-oral photographs of teeth according to the ICDAS guidelines in order to determine the probability of caries. In addition to impacting the practice of dentistry, the system could help with training of dental students’ in diagnose of lesions in early stages.

Method: ,
This study uses computational methods that analyze photographs captured by an intra-oral cameras and produces predictions as to the existence and the severity of caries. The method segments the tooth image into background, healthy enamel surface, and irregular regions (if any),   then from enamel and irregular regions extracts about 120 features based on wavelet and Fourier transforms. Next, principal component analysis is used to reduce the number of features. Result ing features are then fed into logistic regression as a classifier. Seventy three intra-oral photographs of occlusal surfaces of extracted teeth were used. Predicted caries levels were then compared to the levels obtained by ICDAS experts.

Result:,
ICDAS defines 7 codes for different stages of caries  development(0-6). In this pilot study, these 7 codes were collapsed into two classes: 0-2 and 3-6. 10-fold cross validation was used to validate the results, which show 80.8% accuracy, 80.9% specificity, and 80.8% sensitivity.

Conclusion: ,
Preliminary results support the idea of forming an automated caries detection system. While the system needs further improvement and validations using larger datasets, it can be integrated into any computer used in dental clinics practices with a minimal cost. This is a significant advantage over existing systems requiring expensive imaging and external hardware.