Comprehensive Benchmarking of current Deep Learning Models for Mandibular Segmentation and the Influence of Medical Image Quality, Patient Characteristics and Anatomical Region

Comprehensive Benchmarking of current Deep Learning Models for Mandibular Segmentation and the Influence of Medical Image Quality, Patient Characteristics and Anatomical Region - AI Mandible Benchmarking

Status

Active, not recruiting

Phases

Unknown

Study type

Observational

Source

DRKS

Registry ID

DRKS00032736

Enrollment

Registered

2024-04-29

Start date

2024-05-06

Completion date

Unknown

Last updated

2025-04-07

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

Segmentation of the mandible, which is necessary for radiotherapy, computer-assisted surgery or diagnostics.

Interventions

Group 1: In a prospective cross-sectional observational study, current AI models for mandibular segmentation in CT and CBCT images will be included. A balanced and controlled dataset (n=1,000) will be

Eligibility

Sex/Gender

All

Inclusion criteria

Inclusion criteria: - Deep learning model developed within the last 5 years - Deep learning models segments the whole mandible - Returns a label map or mesh model as output

Exclusion criteria

Exclusion criteria: - Model is not based on deep learning - No permission to use the model for this research purpose

Design outcomes

Primary

Measure	Time frame
What is the influence of image properties (slice thickness) on the accuracy of AI-segmented mandibles measured by DSC? For this purpose, the AI segmentation is compared with a gold standard (manual segmentation by two experts).	—

Secondary

Measure	Time frame
- What is the influence of image properties (slice thickness, voxel size XY, sharpness, noise, rotation of mandible in sagittal, axial and coronal direction) on the accuracy of AI-segmented mandibles measured by DSC, NSD, HD95 and MASD? - What impact does the presence of patient characteristics (age, biological sex), bone pathology (fractures, cysts, and the like), osteosynthesis material and dentition (artifacts) have on AI-based segmentation quality measured using DSC, NSD, HD95 and MASD? - Are there differences between CTs and CBCTs on AI-based segmentation quality measured by DSC, NSD, HD95 and MASD? - What is the difference in the AI-based segmentation quality of the mandible in different anatomical regions (Mandibular Condyles, Entrance and exit foramens of IAN, Dentition, Inferior Border of Mandible, Mandibular Body) measured by DSC, NSD, HD95 and MASD?	—

Measure

Time frame

- What is the influence of image properties (slice thickness, voxel size XY, sharpness, noise, rotation of mandible in sagittal, axial and coronal direction) on the accuracy of AI-segmented mandibles measured by DSC, NSD, HD95 and MASD? - What impact does the presence of patient characteristics (age, biological sex), bone pathology (fractures, cysts, and the like), osteosynthesis material and dentition (artifacts) have on AI-based segmentation quality measured using DSC, NSD, HD95 and MASD? - Are there differences between CTs and CBCTs on AI-based segmentation quality measured by DSC, NSD, HD95 and MASD? - What is the difference in the AI-based segmentation quality of the mandible in different anatomical regions (Mandibular Condyles, Entrance and exit foramens of IAN, Dentition, Inferior Border of Mandible, Mandibular Body) measured by DSC, NSD, HD95 and MASD?

—

Countries

Austria, Belgium, China, Czechia, Finland, France, Germany, Netherlands, Switzerland, United States

Contacts

Public ContactBehrus Puladi

Department of Oral and Maxillofacial Surgery & Institute of Medical Informatics, University Hospital RWTH Aachen

bpuladi@ukaachen.de+49 241 80 38389

Outcome results

None listed

Source: DRKS (via WHO ICTRP) · Data processed: Feb 4, 2026