Κατεύθυνση Παιδοδοντιατρική (Κλινικές Ειδικεύσεις)
Βιβλιοθήκη Οδοντιατρικής

2019-02-22

2018

Andrikoula Theofani

Κωνσταντίνος Ουλής, Ομότιμος Καθηγητής, Τμήμα Οδοντιατρικής, Σχολή Επιστημών Υγείας, ΕΚΠΑ
Παναγιώτης Λαγουβάρδος, Ομότιμος Καθηγητής, Τμήμα Οδοντιατρικής, Σχολή Επιστημών Υγείας, ΕΚΠΑ
Ηλίας Μαγκλογιάννης, Αναπληρωτής Καθηγητής, Τμήμα Ψηφιακών Συστημάτων, Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών, Πανεπιστήμιο Πειραιώς

Αυτοματοποιημένος εντοπισμός και διάγνωση μασητικών τερηδόνων (ICDAS II) με αλγόριθμους από ψηφιακές εικόνες ενδοστοματικής κάμερας in vivo

Greek

Automated detection and ICDAS II classification of occlusal caries lesions with algorithms on digital images acquired by an intraoral camera in vivo

Introduction
Numerous techniques have been developed to assess the caries status of occlusal surfaces in an attempt to provide supplementary assistance to the direct visual examination and eliminate the great divergence of decisions found between different practitioners or even the same practitioner when the same lesion is evaluated at different times.

Aim
The aim of this study is to apply and evaluate the reliability of an automated occlusal caries diagnostic system via algorithms on digital images acquired by teeth in vivo using an intraoral camera.

Material and Methods
The examiner was theoretically and practically trained using the standardization tools of the ICDAS II system and calibrated with the golden examiner by examining teeth intraorally and from photographs on the computer screen. Then their findings were compared until they reached an agreement of 98% (>85%).
The sample consisted of 106 permanent molars belonging to patients treated at the postgraduate clinic of the Paediatric Dentistry department and the undergraduate clinic of the University of Athens. Teeth with developmental defects of the enamel or dentin were excluded from the study. Teeth were cleaned of organic residues and stains with a bristle brush, rinsed using air-water spray and dried for 5 seconds using air before every examination and photo shooting. Intraoral camera Carestream - CS 1200 was used for the photo shooting. Then, the examiner assessed the occlusal surfaces in vivo –direct observation- and after classifying each area of interest of the occlusal surface according to ICDAS II, the outline and code of the lesion was marked on a printed image acquired by the camera. The examiner reassessed the same teeth after 15 days from the digital images in the same way on the computer screen –indirect observation-. An application of algorithms developed in Java environment was used for the recording of the direct and indirect examination’s result.
86 of these images (with a number of 3024 superpixels) were used to train the algorithm. Then 20 different images containing multiple occlusal areas of interest were evaluated by the algorithm and the results were compared to the ones from the direct and indirect visual examination.

Statistical analysis
Data were processed and analyzed with STATA 12. The classification from the examiner (after direct and indirect observation) was compared with the classification from the algorithm. Overall proportion of agreement and a weighted and non-weighted version of kappa coefficient of agreement was calculated.

Results
Comparing the system’s classification to the examiner’s classification from the direct observation in the mouth we found the sensitivity ranging from 29.2% to 100% (mean 59.1%), the specificity ranging from 52.4% to 99.2% (mean 91.3%), the precision ranging from 11.1% to 95.4% (mean 49.9%) among the ICDAS II categories and system’s accuracy 74%. Kappa weighted value was calculated to 0.701 and kappa non-weighted value to 0.587.
Comparing the system’s classification to the examiner’s classification from the indirect observation of the photographs we found the sensitivity ranging from 28.6% to 100% (mean 57.3%), the specificity ranging from 56.7% to 98.5% (mean 92.2%), the precision ranging from 8.3% to 97.4% (mean 46.2%) among the ICDAS II categories and system’s accuracy 71.6%. Kappa weighted value was calculated to 0.761 and kappa non-weighted value to 0.601.

Conclusions
The preliminary results indicated that using algorithms for the automated detection and classification of occlusal carious lesions according to ICDAS II system from photographs of an intraoral camera is a very promising and encouraging technique. There are some more adjustments to be made on the training of the algorithm, after which we are expecting to achieve higher numbers of sensitivity and specificity and improve the overall reproducibility and accuracy of the system.

Health Sciences

Caries, Occlusal, Diagnosis, Algorithm, Photograph

No

0

Yes

175

94

File access is restricted only to the intranet of UoA.

https://pergamos.lib.uoa.gr/uoa/dl/object/2864011