Fig 5 - uploaded by Haofu Liao
Content may be subject to copyright.
ROC curve of cavity detection by the SSD-MobileNetV2 model. The red box denotes the average performance of dentists. The curve does not reach the upper right corner because we take into account the undetected teeth and incorrect detections.
Source publication
Early childhood caries (ECC) is the most common, yet preventable chronic disease in children under the age of 6. Treatments on severe ECC are extremely expensive and unaffordable for socioeconomically disadvantaged families. The identification of ECC in an early stage usually requires expertise in the field, and hence is often ignored by parents. T...
Context in source publication
Context 1
... extra row and column take into account the undetected teeth and incorrect detections (detected unlabeled areas). We then plot the ROC curve in Fig.5 by calculating from the confusion matrix the sensitivity and specificity scores with respect to cavity detection instead of per category detection. ...
Citations
... In addition, the Alliance for a Cavity-Free Future states that around 530 million children have untreated tooth cavities. This could affect their future permanent teeth or need expensive surgery if not noticed or treated well [9] . Therefore, a more applicable solution will be improved by training a deep learning model to detect cavities on real-life images of children's teeth. ...
... Convolutional neural networks and deep learning have recently been used in the first studies to detect caries on dental X-rays [11,1]. However, multiple efforts have lately been made to detect tooth decay in intraoral images using Artificial Intelligence [4,11,9,12] . Srivastava et al. [1] developed a computer-aided diagnosis (CAD) system that improves dentists' abilities to identify a variety of dental cavities from bitewing radiographs. ...
... The study hypothesizes that future work should include additional and correctly segmented images to improve results. Zhang et al. [9] also used a MobileNetV2 backbone, but it was used for a Single Shot Mobile Detector (SSD). This model's performance was compared to a Faster Region-Based Convolutional Neural Network (Faster R-CNN) using a ResNet50 backbone and dentists. ...
Approximately 530 million children have untreated cavities, which could affect future permanent teeth if not treated well. Researchers have enhanced X-ray detection of tooth decay to improve the detection of cavities. However, dentophobia and lack of insurance and dentist availability are barriers that constrain thousands from receiving proper dental care. This study used a deep learning convolutional neural network model to address this problem to detect cavities. Three hundred twenty-two photos taken by a camera of child patients were used. The dataset was classified into two classes: cavity and no cavity. The MobileNetV2 architecture was used for feature extraction and cavity detection. The model was then trained and tested so that it classified photos into two categories. The model achieved an accuracy of 93.5%.
... Recently, AI had been tested in detecting caries and oral pathologies on dental x-rays [37,38]. Our team has developed a smartphone app, AICaries, that uses AI-powered technology to detect caries on photographs of teeth taken via smartphone [39][40][41]. As AI is currently used to aid imaging recognition to improve disease diagnosis in many medical fields, including oncology, ophthalmology, and radiology [42][43][44][45], AI has the full potential to be developed in dentistry for remote caries detection and caries risk management for underserved patients with limited access to oral health care. ...
Background:
Amid the COVID-19 pandemic and other possible future infectious disease pandemics, dentistry needs to consider modified dental examination regimens that render quality care and ensure the safety of patients and dental health care personnel (DHCP).
Objective:
This study aims to assess the acceptance and usability of an innovative mDentistry eHygiene model amid the COVID-19 pandemic.
Methods:
This pilot study used a 2-stage implementation design to assess 2 critical components of an innovative mDentistry eHygiene model: virtual hygiene examination (eHygiene) and patient self-taken intraoral images (SELFIE), within the National Dental Practice-Based Research Network. Mixed methods (quantitative and qualitative) were used to assess the acceptance and usability of the eHygiene model.
Results:
A total of 85 patients and 18 DHCP participated in the study. Overall, the eHygiene model was well accepted by patients (System Usability Scale [SUS] score: mean 70.0, SD 23.7) and moderately accepted by dentists (SUS score: mean 51.3, SD 15.9) and hygienists (SUS score: mean 57.1, SD 23.8). Dentists and patients had good communication during the eHygiene examination, as assessed using the Dentist-Patient Communication scale. In the SELFIE session, patients completed tasks with minimum challenges and obtained diagnostic intraoral photos. Patients and DHCP suggested that although eHygiene has the potential to improve oral health care services, it should be used selectively depending on patients' conditions.
Conclusions:
The study results showed promise for the 2 components of the eHygiene model. eHygiene offers a complementary modality for oral health data collection and examination in dental offices, which would be particularly useful during an infectious disease outbreak. In addition, patients being able to capture critical oral health data in their home could facilitate dental treatment triage and oral health self-monitoring and potentially trigger oral health-promoting behaviors.
... On the Caries Status Report interface (Fig 1C), user can choose "Reduce Risk" to access Perinatal Oral Health Education (Fig 1F) or click "Find a dentist" for dental clinic information. The performance of AICaries in caries detection was reported previously [17]. ...
Early Childhood Caries (ECC) is the most common childhood disease worldwide and a health disparity among underserved children. ECC is preventable and reversible if detected early. However, many children from low-income families encounter barriers to dental care. An at-home caries detection technology could potentially improve access to dental care regardless of patients’ economic status and address the overwhelming prevalence of ECC. Our team has developed a smartphone application (app), AICaries, that uses artificial intelligence (AI)-powered technology to detect caries using children’s teeth photos. We used mixed methods to assess the acceptance, usability, and feasibility of the AICaries app among underserved parent-child dyads. We conducted moderated usability testing (Step 1) with ten parent-child dyads using "Think-aloud" methods to assess the flow and functionality of the app and analyze the data to refine the app and procedures. Next, we conducted unmoderated field testing (Step 2) with 32 parent-child dyads to test the app within their natural environment (home) over two weeks. We administered the System Usability Scale (SUS) and conducted semi-structured individual interviews with parents and conducted thematic analyses. AICaries app received a 78.4 SUS score from the participants, indicating an excellent acceptance. Notably, the majority (78.5%) of parent-taken photos of children’s teeth were satisfactory in quality for detection of caries using the AI app. Parents suggested using community health workers to provide training to parents needing assistance in taking high quality photos of their young child’s teeth. Perceived benefits from using the AICaries app include convenient at-home caries screening, informative on caries risk and education, and engaging family members. Data from this study support future clinical trial that evaluates the real-world impact of using this innovative smartphone app on early detection and prevention of ECC among low-income children.
Objectives:
This systematic review aimed at evaluating the performance of artificial intelligence (AI) models in detecting dental caries on oral photographs.
Methods:
Methodological characteristics and performance metrics of clinical studies reporting on deep learning and other machine learning algorithms were assessed. The risk of bias was evaluated using the quality assessment of diagnostic accuracy studies 2 (QUADAS-2) tool. A systematic search was conducted in EMBASE, Medline, and Scopus.
Results:
Out of 3410 identified records, 19 studies were included with six and seven studies having low risk of biases and applicability concerns for all the domains, respectively. Metrics varied widely and were assessed on multiple levels. F1-scores for classification and detection tasks were 68.3%-94.3% and 42.8%-95.4%, respectively. Irrespective of the task, F1-scores were 68.3%-95.4% for professional cameras, 78.8%-87.6%, for intraoral cameras, and 42.8%-80% for smartphone cameras. Limited studies allowed assessing AI performance for lesions of different severity.
Conclusion:
Automatic detection of dental caries using AI may provide objective verification of clinicians' diagnoses and facilitate patient-clinician communication and teledentistry. Future studies should consider more robust study designs, employ comparable and standardized metrics, and focus on the severity of caries lesions.
