Page 88 - AIH-1-3

Artificial Intelligence in Health Dental cavity prediction with computer vision

segmentation techniques into the research framework. precision, recall, and mean average precision (mAP). The
Four distinct segmentation methods were evaluated, YOLO model ensemble achieved a notable mAP of 0.732,
namely segmentation with the thresholding method, an accuracy of 0.789, and a recall of 0.701. When applied
segmentation with the contouring method, segmentation to VGG16, the final model demonstrated impressive
with the Canny-edges method, and segmentation with a diagnostic accuracy of 86.96%, with precision and recall
combination of these techniques. The best performance values of 0.89 and 0.88, respectively. This performance
among all methods was obtained by the Canny edge-CNN outstripped all other existing methods for object detection
mode. 6 in free-hand, non-standardized smartphone photographs.
In response to the inefficiency and complexity of The virtual computer vision AI system, enriched by an
traditional dental disease detection methods, a study ensemble model, test-time augmentation, and transfer
introduced a novel approach utilizing deep learning. learning techniques, successfully predicts dental
7
The study employed the YOLOv3 model to automate cavitations from non-standardized photographs with
the detection and classification of four common teeth clinically reasonable accuracy. This innovation holds the
problems: cavities, root canals, dental crowns, and broken- potential to enhance access to oral health care in resource-
down root canals, using panoramic dental X-ray images constrained, underserved areas and facilitates automated
orthopantograms. To overcome data limitations, a dental diagnostics and advanced tele-dentistry applications.
X-ray dataset with 1200 augmented images was created Thanh et al., demonstrated the potential of mobile
11
and divided into 70% for training and 30% for testing. The phone-based diagnostic tools for dental caries detection
YOLOv3 model achieved a remarkable 99.33% accuracy, using deep learning algorithms, highlighting the efficiency
outperforming existing models and demonstrating its of YOLOv3 and Faster R-CNN models. A blog article
versatility with other datasets. 7 on innovative applications in dentistry showcased AIs
12
Another study indicated that deep learning models ability to detect caries with high accuracy using image
can be used to help dentists in planning dental implant augmentation and transfer learning, emphasizing its role
placement, ensuring that dental implants are optimally in complementing traditional diagnostic methods. In
placed and properly aligned with the surrounding teeth addition, a GitHub project has been established, aiming
and bone. 8 to detect and localize various dental diseases, including
caries and periodontal diseases, using computer vision in
Dental caries, one of the most prevalent dental conditions panoramic dental X-ray images. 13
in contemporary times, poses significant challenges for
14
early detection in dental X-ray or radiovisiography images. Nakai and Wei, while focusing on protein localization,
Deep learning has been widely employed across medical highlighted the adaptability of deep learning techniques,
domains for predictive and diagnostic purposes. One of such as CNN and long short-term memory, for predictive
the investigations evaluated a K-means clustering approach modeling across diverse fields, including dentistry.
15
for image segmentation, underscoring the significance Acharya discussed deep learning techniques for image
of image enhancement techniques in improving the segmentation, including U-Net and SegNet, which are
quality of dental radiographs. The implemented K-means crucial for detailed analysis in medical imaging and
model algorithm demonstrated improved accuracy in the diagnostics. Brownlee explored the architectures of Fast
16
detection of dental caries. 9 R-CNN and Faster R-CNN for real-time object detection,
relevant for precise localization and quantification in
Tareq et al., aimed to pioneer a novel and cost- 17
10
effective virtual computer vision artificial intelligence (AI) dental imaging. Fernandes et al., while focused on animal
sciences, underscored the importance of machine learning
system capable of predicting dental cavitation from non- and deep learning algorithms in various computer vision
standardized photographs with reasonable clinical accuracy.
They curated a dataset comprising 1703 augmented images applications, illustrating the multidisciplinary potential of
sourced from 233 de-identified teeth specimens, captured these technologies.
using consumer-grade smartphones. The methodology A study on gait pattern recognition for flat fall prediction
leveraged cutting-edge techniques, including ensemble highlighted the use of computer vision and machine
modeling, test-time augmentation, and transfer learning learning in recognizing gait patterns, demonstrating the
processes. The researchers independently assessed versatility of these technologies in health diagnostics
derivatives of the YOLO algorithm, including v5s, v5m, beyond dental applications. A notable study utilized
18
v5l, and v5x, subsequently creating an ensemble model and CNNs to diagnose dental caries from bitewing images,
transfer-learning it with ResNet50, ResNet101, VGG16, emphasizing the complexity of identifying proximal
AlexNet, and DenseNet. Evaluation metrics encompass and interproximal dental caries and the effectiveness

Volume 1 Issue 3 (2024) 82 doi: 10.36922/aih.3184