Page 87 - AIH-1-3

Artificial Intelligence in Health Dental cavity prediction with computer vision

However, some of these bacteria have the potential to cause model dedicated to the precise localized identification
health problems, as they serve as a gateway to our digestive of dental cavities within colored images. We recognize
and respiratory systems. that the accurate identification of cavity locations is an
The body’s own defense typically keeps these germs indispensable initial step in streamlining the diagnostic
in check, along with basic dental hygiene habits like process. Achieving this objective will empower dental
frequent brushing and flossing. However, without proper professionals with a powerful tool that not only identifies
oral hygiene, bacteria levels can rise and cause ailments cavities but also provides their exact spatial coordinates.
such as tooth decay and gum disease. Research indicates Building upon the success of our initial objective, our
that certain medical conditions may be impacted by oral second key goal is to introduce a methodology for the
bacteria and the inflammation brought on by chronic gum quantification of dental cavity areas within these localized
disease, also known as periodontitis. Our oral health might regions. Accurate area quantification is paramount in
contribute to various diseases and conditions, including assessing the severity of cavities and can significantly
endocarditis, cardiovascular disease, and pneumonia. 1 aid in treatment planning. With this objective, we aim
According to the Global Burden studies in 2019, dental to automate and standardize the area measurement
caries is the most common oral disease, affecting around process, thereby enhancing the precision of dental health
3.5 billion people, of whom 2 billion have permanent dental assessments. Furthermore, it is noteworthy to establish a
caries. Moreover, 1.45 million of the 6 million patients subtle connection between our current study’s objectives
2
with dental caries who visited dentists in the Republic of and our prior research endeavors. In our earlier work, we
Korea in 2020 were children (0 – 9 years old). Therefore, it endeavored to forecast dental cavities utilizing convolutional
3
is important to study tooth caries. neural networks (CNNs). While that research focused
on predicting the occurrence of cavities, the objectives
Caries formation is affected by a host of preferential of the current study extend beyond prediction. Here, we
habits, systemic disorders, and congenital anomalies. embark on the critical task of geospatial mapping them
Incipient carious lesions are frequently overlooked within images, quantifying their extent, and subsequently
by patients. Dentists treat them conservatively using facilitating a more comprehensive understanding of their
techniques of minimally invasive dentistry. As a result, impact on oral health. This intrinsic linkage between our
there are many instances of misdiagnosis and poor care, research pursuits underscores the holistic approach we
particularly among young practitioners conducting visual adopt in addressing the multifaceted challenges in dental
and radiographic investigations. Caries mismanagement cavity analysis, paving the way for a more comprehensive
can be expensive and leave the patient exposed to future and integrated solution.
periapical, osseous, and fascial space spread of the infection.
4
These challenges are particularly noteworthy in areas with In this study, we collected a dataset of colored images
limited access to advanced dental facilities and trained containing dental cavities and manually annotated this
practitioners. Early detection and ongoing monitoring of dataset using Roboflow annotation. We then trained the
these problems are proposed in this article using a low- “You Only Look Once” (YOLO) v5 model to detect and
cost automated system that does not differentiate patients locate dental cavities in these images using a bounding
based on their sociodemographic status. box. Once we identified the exact cavity area with a
bounding box, we used the coordinates of the bounding
Since cellular technology has expanded globally, even box to calculate the area of the cavity. Applying image
to rural areas, the use of mobile portable devices like segmentation on the cavity highlighted the cavity area, and
smartphones has increased exponentially in emerging cavity masks were obtained from segmentation for further
economies. As a result, biomedical research can harness analysis of the dental cavity shape.
5
functionality in smartphones to offer cost-effective
solutions to challenging issues in oral treatment. 2. Literature review
In this research endeavor, we define several core In our prior research endeavor, entitled “Forecasting
objectives, each serving as a pivotal milestone in our pursuit teeth cavities by CNNs,” we conducted a comprehensive
of advancing dental cavity analysis through computer exploration into predicting dental cavities using CNNs.
vision techniques. These objectives collectively constitute The dataset in this previous investigation consisted of
the foundation upon which our study is constructed, X-ray images. To augment the dataset’s size and diversify its
steering our research toward meaningful and innovative content, we applied a series of sophisticated augmentation
contributions to the field. Our first and foremost objective techniques. To further enhance the accuracy and efficacy
is to pioneer the development of a robust computer vision of dental cavity prediction, we methodically incorporated

Volume 1 Issue 3 (2024) 81 doi: 10.36922/aih.3184