Overcoming Challenges in Dental Deep Learning: Presentation Insights

This presentation by Martha Büttner at the AI for Dentistry Symposium delves into current challenges in dental deep learning, highlighting issues like data sharing, annotation bottlenecks, and comparability gaps. The talk proposes a solution through Federated Learning, showcasing a project on Tooth Segmentation using data from multiple centers. By addressing data sharing barriers and leveraging cross-center models, the presentation demonstrates how Federated Learning outperforms local learning methods across various centers, emphasizing the importance of generalizability in dental AI models.

• Dental Deep Learning
• AI for Dentistry
• Federated Learning
• Data Sharing
• Tooth Segmentation

rahe_i Follow

Uploaded on Sep 26, 2024 | 0 Views

Download Presentation

Please find below an Image/Link to download the presentation.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.

Presentation Transcript

1. FGAI4H-R-040-A06 Cambridge, 21-24 March 2023 Source: Charit Universit tsmedizin Berlin Title: Att.6 Presentation - Overcoming Current Challenges in Dental Deep Learning Contact: Martha B ttner E-mail: martha.buettner@charite.de Abstract: This PPT contains a presentation on overcoming current challenges in dental deep learning given in the AI for Dentistry Symposium on 21 March 2023.

2. Overcoming Current Challenges in Dental Deep Learning Martha B ttner | 21.03.2023 | Dental Symposium WHO & ITU FG AI4H Oral Diagnostics, Digital Dentistry and Health Services Research

3. Agenda - Challenges 1. Data sharing 2. Annotation bottle neck 3. Missing comparability 3 Oral Diagnostics, Digital Dentistry and Health Services Research

4. 1 Data Sharing 4 Oral Diagnostics, Digital Dentistry and Health Services Research

5. Challenge International Data Sharing AI requires big amount of data Medical data is high sensitive Especially dental image data difficult to de-identify Data protection barriers are high Generalizabilty of deep learning models of high importance AI model performance differ when exposed to data from different centers Leeds to bias and unfair medical diagnostic 5 Oral Diagnostics, Digital Dentistry and Health Services Research

6. Proposed Solution Federated Learning model on cross-center data without sharing 6 Oral Diagnostics, Digital Dentistry and Health Services Research

7. Project Federated Learning for Tooth Segmentation Simulation of FL on data from 9 different centers Tooth segmentation on panoramic images (n=143 to n=1,881 per center) Compared against local learning 8 out of 9 centers: FL outperformed local learning FL outperformed local learning across all centers in generalizability 7 Oral Diagnostics, Digital Dentistry and Health Services Research

8. 2 Annotation Bottle Neck 8 Oral Diagnostics, Digital Dentistry and Health Services Research

9. Challenge Medical Annotation Most medical AI solution trained in a supervised manner High amount of labeled data required Expert needed for annotation Time consuming Cost intensive 9 Oral Diagnostics, Digital Dentistry and Health Services Research

10. Proposed Solution Semi Supervised Learning First Model trained on small amount of labeled data Teacher model Prediction on unlabeled data Prediction used to train a new model Student model Fine tuning on labeled data Student model becomes a teacher model 10 Oral Diagnostics, Digital Dentistry and Health Services Research

11. Project Semi supervised caries segmentation Application of semi-supervised learning to two diffrent problems and data types Angle classification on intraoral photographs Segmentation of caries lesions on bitewing radiographs Model benefited from semi supervised approach Student model outperformed teacher models significantly (evaluated metrics: Dice, IoU, Sensitivity, PPV) 11 Oral Diagnostics, Digital Dentistry and Health Services Research

12. 3 Missing Comparability 12 Oral Diagnostics, Digital Dentistry and Health Services Research

13. Challenge Missing Comparability Inconsistent reporting Systematic review identified high amount of different metrics Difficult to perform meta analysis 13 Oral Diagnostics, Digital Dentistry and Health Services Research

14. Challenge Inconsistent reporting Systematic review identified immensive amount of diffrent metrics Difficult to perform meta analysis Current reporting guidelines do not focus on metircs Metrics need technical knowledge to interpret Medical devices need evaluation on clinical metrics based on confusion matrix 14 Oral Diagnostics, Digital Dentistry and Health Services Research

15. Challenge Missing Comparability Inconsistent reporting Systematic review identified high amount of different metrics Difficult to perform meta analysis Current reporting guidelines do not focus on metrics Metrics need technical knowledge to interpret Medical devices need evaluation on clinical metrics based on confusion matrix 15 Oral Diagnostics, Digital Dentistry and Health Services Research

16. Outlook Core Outcome Development Development of reporting guidelines for dental computer vision Combining clinical and technical perspective Translation to clinical interpretable metrics Consensus for reporting requirement 1. Manuals for generating clinical relevant metrics 2. 16 Oral Diagnostics, Digital Dentistry and Health Services Research

17. Thank you for your attention! Martha B ttner | martha.buettner@charite.de Oral Diagnostics, Digital Dentistry and Health Services Research