Deep Learning to Summarize Findings in Dental Panoramic Radiographs

• Conditions: Digital; Dentist-Patient Relations; Radiography; Dental Diseases
• Interventions: Radiation: Dental Panoramic Radiographs

• Registration Number: NCT04894201
• Lead Sponsor: National Taiwan University Hospital

Brief Summary

In this work, the investigators study the application of artificial intelligence systems on dental panoramic images for dental findings. An artificial intelligence system will be learned on an publicly available panoramic image dataset, and test against the investigators' local patient cohort as external test data. The investigators hypothesize the performance would be similar, if not identical to on the public data, and that the investigators' AI system is generalizable.

Detailed Description

Not available

Study Timeline

• Status Verified: 2021-05
• Study Start: 2021-05-07
• Study First Submitted: 2021-05-09
• Study First Submitted QC: 2021-05-16
• Last Update Submitted: 2021-05-16
• Study First Posted: 2021-05-20
• Last Update Posted: 2021-05-20
• Primary Completion: 2021-12
• Study Completion: 2023-05

Recruitment & Eligibility

• Status: UNKNOWN
• Sex: All
• Target Recruitment: 1000

Inclusion Criteria

Not provided

Read More

Exclusion Criteria

• patients under 20 or with primary teeth
• patients with non-removable metal accessory above neck, such as tongue ring, nose ring etc.
• patients with mandible or maxilla deformation

Read More

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Arm && Interventions

Group	Intervention	Description
All patients with relevant dental findings	Dental Panoramic Radiographs	-

Primary Outcome Measures

Name	Time	Method
Area Under Curve for Receiver Operating Characteristics of Clinical Findings	1 day	The investigators use the AI model to infer whether a clinical finding (out of the six type of findings the investigators are interested in) is present in an imaging study from the test set. The result is compared against expert annotation and evaluated for receiver operating characteristics over the whole set. The area under curve will then be calculated and averaged across six type of findings to represent the overall efficacy of the model on detecting findings from panoramic images. From a scale of zero to one, zero is the worst this metric can be and one is the best.

Trial Locations

Locations (1)

National Taiwan University Hospital; 🇨🇳 Taipei, Taiwan