PhD Scientific Days 2026

Poster Session 1.O - Dental Research

Extra- and Intracranially Located Carotid Plaque Research on CBCT Images Supported by Artificial Intelligence

Name of the presenter

Kárpáti, Márton Tamás

Institute/workplace of the presenter

Department of Oral Diagnostics, Semmelweis University

Authors

Márton Tamás Kárpáti¹
1: Department of Oral Diagnostics, Semmelweis University

Text of the abstract

Introduction

Cardiovascular diseases belong to the most common illnesses all over the world. Atherosclerosis is the part of them and affect the carotid arteries in the head and neck region. The entity is called carotid plaque, which can be located extra-, and intracranial and could potentially lead to life-threatening issues. During dental radiographic imaging the calcified carotid plaque could be detected as an incidental finding on the Cone-Beam Computed Tomography (CBCT) imaging method. The improvement of the accuracy for carotid calcification detection can be supported by convolutional neural network based artificial intelligence algorithms.

Aims

Our project aims to assess a deep learning-based software environment on reconstructed CBCT images, which can enhance the accuracy and precision of extra- and intracranial carotid artery calcification detection.

Method

During our research 2766 CBCT images were examined retrospectively which were made at the Department of Oral Diagnostics between 01.06.2019 and 31.05.2023. The CBCT data were evaluated by three observers. After the collection phase we manually segmented and annotated the reported entities, then the files we imported into the deep learning based convolutional neural network (Ankara University, Faculty of Dentistry, Oral and Maxillofacial Radiology Department). We used a 3D-ResNet based hybrid model.

Results

In total we detected calcified carotid plaques on 97 CBCT images. 113 entities were located extracranial and 111 were found intracranial on the CBCT radiographs. The examined deep learning-based 3D-ResNet model classified the selected CBCT data, and the efficiency were evaluated by the values of AUC 0.93, sensitivity 0.92, specificity 0.95 and precision 0.93.

Conclusion

The 3D-ResNet algorithm executed successfully during the research process, and it showed progressive efficiency values for extra- and intracranial carotid artery classification and detection as well. Consequently, deep learning based neural network can help detecting the potentially life-threatening findings beyond dentoalveolar region as decision support systems.

Funding

Supported by the 2025-2.1.1-EKÖP-2025-00014 University Research Scholarship Programme of the Ministry for Culture and Innovation from the source of the National Research, Development and Innovation Fund.