Automated detection and classification of osteolytic lesions in panoramic radiographs using CNNs and vision transformers
Abstract Background Diseases underlying osteolytic lesions in jaws are characterized by the absorption of bone tissue and are often asymptomatic, delaying their diagnosis. Well-defined lesions (benign cyst-like lesions) and ill-defined lesions (osteomyelitis or malignancy) can be detected early in a...
Saved in:
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2025-06-01
|
| Series: | BMC Oral Health |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s12903-025-06209-6 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849329378669887488 |
|---|---|
| author | Niels van Nistelrooij Iman Ghanad Amir K. Bigdeli Daniel G. E. Thiem Constantin von See Carsten Rendenbach Ira Maistreli Tong Xi Stefaan Bergé Max Heiland Shankeeth Vinayahalingam Robert Gaudin |
| author_facet | Niels van Nistelrooij Iman Ghanad Amir K. Bigdeli Daniel G. E. Thiem Constantin von See Carsten Rendenbach Ira Maistreli Tong Xi Stefaan Bergé Max Heiland Shankeeth Vinayahalingam Robert Gaudin |
| author_sort | Niels van Nistelrooij |
| collection | DOAJ |
| description | Abstract Background Diseases underlying osteolytic lesions in jaws are characterized by the absorption of bone tissue and are often asymptomatic, delaying their diagnosis. Well-defined lesions (benign cyst-like lesions) and ill-defined lesions (osteomyelitis or malignancy) can be detected early in a panoramic radiograph (PR) by an experienced examiner, but most dentists lack appropriate training. To support dentists, this study aimed to develop and evaluate deep learning models for the detection of osteolytic lesions in PRs. Methods A dataset of 676 PRs (165 well-defined, 181 ill-defined, 330 control) was collected from the Department of Oral and Maxillofacial Surgery at Charité Berlin, Germany. The osteolytic lesions were pixel-wise segmented and labeled as well-defined or ill-defined. Four model architectures for instance segmentation (Mask R-CNN with a Swin-Tiny or ResNet-50 backbone, Mask DINO, and YOLOv5) were employed with five-fold cross-validation. Their effectiveness was evaluated with sensitivity, specificity, F1-score, and AUC and failure cases were shown. Results Mask R-CNN with a Swin-Tiny backbone was most effective (well-defined F1 = 0.784, AUC = 0.881; ill-defined F1 = 0.904, AUC = 0.971) and the model architectures including vision transformer components were more effective than those without. Model mistakes were observed around the maxillary sinus, at tooth extraction sites, and for radiolucent bands. Conclusions Promising deep learning models were developed for the detection of osteolytic lesions in PRs, particularly those with vision transformer components (Mask R-CNN with Swin-Tiny and Mask DINO). These results underline the potential of vision transformers for enhancing the automated detection of osteolytic lesions, offering a significant improvement over traditional deep learning models. |
| format | Article |
| id | doaj-art-0b2dd6e9fbbf4d898ab96ecba1a14acf |
| institution | Kabale University |
| issn | 1472-6831 |
| language | English |
| publishDate | 2025-06-01 |
| publisher | BMC |
| record_format | Article |
| series | BMC Oral Health |
| spelling | doaj-art-0b2dd6e9fbbf4d898ab96ecba1a14acf2025-08-20T03:47:17ZengBMCBMC Oral Health1472-68312025-06-0125111010.1186/s12903-025-06209-6Automated detection and classification of osteolytic lesions in panoramic radiographs using CNNs and vision transformersNiels van Nistelrooij0Iman Ghanad1Amir K. Bigdeli2Daniel G. E. Thiem3Constantin von See4Carsten Rendenbach5Ira Maistreli6Tong Xi7Stefaan Bergé8Max Heiland9Shankeeth Vinayahalingam10Robert Gaudin11Department of Oral and Maxillofacial Surgery, Radboud University Medical CenterDepartment of Oral and Maxillofacial Surgery, Charité - Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt Universität zu BerlinDepartment of Hand, Plastic and Reconstructive Surgery, BG Trauma Center Ludwigshafen, University of HeidelbergDepartment of Oral and Maxillofacial Surgery, University Medical Centre, Johannes Gutenberg University MainzDepartment of Dentistry, Faculty of Medicine and Dentistry, Danube Private UniversityDepartment of Oral and Maxillofacial Surgery, Charité - Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt Universität zu BerlinDepartment of Oral and Maxillofacial Surgery, Charité - Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt Universität zu BerlinDepartment of Oral and Maxillofacial Surgery, Radboud University Medical CenterDepartment of Oral and Maxillofacial Surgery, Radboud University Medical CenterDepartment of Oral and Maxillofacial Surgery, Charité - Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt Universität zu BerlinDepartment of Oral and Maxillofacial Surgery, Radboud University Medical CenterDepartment of Oral and Maxillofacial Surgery, Charité - Universitätsmedizin Berlin, corporate member of Freie Universität Berlin and Humboldt Universität zu BerlinAbstract Background Diseases underlying osteolytic lesions in jaws are characterized by the absorption of bone tissue and are often asymptomatic, delaying their diagnosis. Well-defined lesions (benign cyst-like lesions) and ill-defined lesions (osteomyelitis or malignancy) can be detected early in a panoramic radiograph (PR) by an experienced examiner, but most dentists lack appropriate training. To support dentists, this study aimed to develop and evaluate deep learning models for the detection of osteolytic lesions in PRs. Methods A dataset of 676 PRs (165 well-defined, 181 ill-defined, 330 control) was collected from the Department of Oral and Maxillofacial Surgery at Charité Berlin, Germany. The osteolytic lesions were pixel-wise segmented and labeled as well-defined or ill-defined. Four model architectures for instance segmentation (Mask R-CNN with a Swin-Tiny or ResNet-50 backbone, Mask DINO, and YOLOv5) were employed with five-fold cross-validation. Their effectiveness was evaluated with sensitivity, specificity, F1-score, and AUC and failure cases were shown. Results Mask R-CNN with a Swin-Tiny backbone was most effective (well-defined F1 = 0.784, AUC = 0.881; ill-defined F1 = 0.904, AUC = 0.971) and the model architectures including vision transformer components were more effective than those without. Model mistakes were observed around the maxillary sinus, at tooth extraction sites, and for radiolucent bands. Conclusions Promising deep learning models were developed for the detection of osteolytic lesions in PRs, particularly those with vision transformer components (Mask R-CNN with Swin-Tiny and Mask DINO). These results underline the potential of vision transformers for enhancing the automated detection of osteolytic lesions, offering a significant improvement over traditional deep learning models.https://doi.org/10.1186/s12903-025-06209-6Deep LearningOsteolytic LesionsPanoramic RadiographVision Transformer |
| spellingShingle | Niels van Nistelrooij Iman Ghanad Amir K. Bigdeli Daniel G. E. Thiem Constantin von See Carsten Rendenbach Ira Maistreli Tong Xi Stefaan Bergé Max Heiland Shankeeth Vinayahalingam Robert Gaudin Automated detection and classification of osteolytic lesions in panoramic radiographs using CNNs and vision transformers BMC Oral Health Deep Learning Osteolytic Lesions Panoramic Radiograph Vision Transformer |
| title | Automated detection and classification of osteolytic lesions in panoramic radiographs using CNNs and vision transformers |
| title_full | Automated detection and classification of osteolytic lesions in panoramic radiographs using CNNs and vision transformers |
| title_fullStr | Automated detection and classification of osteolytic lesions in panoramic radiographs using CNNs and vision transformers |
| title_full_unstemmed | Automated detection and classification of osteolytic lesions in panoramic radiographs using CNNs and vision transformers |
| title_short | Automated detection and classification of osteolytic lesions in panoramic radiographs using CNNs and vision transformers |
| title_sort | automated detection and classification of osteolytic lesions in panoramic radiographs using cnns and vision transformers |
| topic | Deep Learning Osteolytic Lesions Panoramic Radiograph Vision Transformer |
| url | https://doi.org/10.1186/s12903-025-06209-6 |
| work_keys_str_mv | AT nielsvannistelrooij automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT imanghanad automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT amirkbigdeli automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT danielgethiem automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT constantinvonsee automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT carstenrendenbach automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT iramaistreli automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT tongxi automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT stefaanberge automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT maxheiland automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT shankeethvinayahalingam automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers AT robertgaudin automateddetectionandclassificationofosteolyticlesionsinpanoramicradiographsusingcnnsandvisiontransformers |