Abstract
Amaç: Bu çalışmada, dental panoramik radyografilerde periapikal lezyonların yapay zekâ destekli otomatik tespiti amacıyla iki farklı derin öğrenme modeli olan YOLOv8 ve Cascade Mask R-CNN’nin performansları karşılaştırılmıştır.
Yöntem: Toplamda 1861 panoramik radyografi incelenmiş ve periapikal lezyon barındıran 602 görüntü çalışma kapsamına alınmıştır. Bu radyograflarda toplamda 1059 lezyonun bulunduğu belirlenmiştir. Lezyon tespiti için gözlemciler arası uyum Kappa analizi ile değerlendirilmiştir. Çalışmaya dahil edilen radyograflar, eğitim (%70), doğrulama (%20) ve test (%10) olmak üzere üç veri setine ayrılmıştır. YOLOv8 ve Cascade Mask R-CNN yapay zeka modelleri PyTorch kütüphanesi kullanılarak eğitilmiştir. Eğitilen modeller test veri setine uygulanmış ve periapikal lezyonlar için gerçek pozitif, yanlış pozitif, gerçek negatif ve yanlış negatif durumları tespit edilmiştir. Bu bulgular üzerinden ise modellere ait doğruluk, duyarlılık, özgüllük, kesinlik ve F1 skoru hesaplanmıştır.
Bulgular: Test veri setinde YOLOv8 modeli %96,9 doğruluk, %53,5 duyarlılık, %90 kesinlik ve 0,671 F1 skoru sunarken; Cascade Mask R-CNN modeli %97,6 doğruluk, %76,2 duyarlılık, %81,9 kesinlik ve 0,790 F1 skoruna ulaşmıştır. YOLOv8 daha az yanlış pozitif üretse de yüksek sayıda lezyonu gözden kaçırmıştır. Cascade Mask R-CNN ise daha fazla lezyonu tespit edebilmiş, ancak nispeten daha fazla yanlış pozitif üretmiştir.
Sonuç: Yapay zeka modelleri, doğruluk oranları dikkate alındığında klinik karar destek sistemlerinde yardımcı araçlar olarak potansiyel taşımaktadır. Ancak mevcut duyarlılık oranlarının tanı sürecini güvenle üstlenebilecek düzeyde olmadığı, bu nedenle bu modellerin tek başlarına klinik karar aracı olarak kullanılmaması gerektiği sonucuna varılmıştır.
References
- 1. Figdor D. Apical periodontitis: a very prevalent problem. Oral Surg Oral Med Oral Pathol 2002;94:651-652.
- 2. Bender I, Seltzer S. Roentgenographic and direct observation of experimental lesions in bone: I. J Am Dent Assoc 1961;62:152-160.
- 3. Bender I, Seltzer S. Roentgenographic and direct observation of experimental lesions in bone: II. J Am Dent Assoc 1961;62:708-716.
- 4. Jorge EG, Tanomaru-Filho M, Gonçalves M, Tanomaru JM. Detection of periapical lesion development by conventional radiography or computed tomography. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2008;106:56-61.
- 5. Dutra KL, Haas L, Porporatti AL, Flores-Mir C, Santos JN, Mezzomo LA, Correa M, Canto GDL. Diagnostic accuracy of cone-beam computed tomography and conventional radiography on apical periodontitis: a systematic review and meta-analysis. J Endod 2016;42:356-364.
- 6. Stera G, Giusti M, Magnini A, Calistri L, Izzetti R, Nardi C. Diagnostic accuracy of periapical radiography and panoramic radiography in the detection of apical periodontitis: a systematic review and meta-analysis. Radiol Med 2024;129:1682-1695.
- 7. Szabó V, Orhan K, Dobó-Nagy C, Veres DS, Manulis D, Ezhov M, Sanders A, Szabó BT. Deep Learning-Based Periapical Lesion Detection on Panoramic Radiographs. Diagnostics 2025;15:510.
- 8. Ba-Hattab R, Barhom N, Osman SAA, Naceur I, Odeh A, Asad A, Al-Najdi SARN, Ameri E, Daer A, Silva RLBD, Costa C, Cortes ARG, Tamimi F. Detection of Periapical Lesions on Panoramic Radiographs Using Deep Learning. Appl Sci 2023;13:1516.
- 9. Landis JR, Koch GG. The Measurement of Observer Agreement for Categorical Data. Biometrics 1977;33:159-174.
- 10. Bayrakdar IS, Orhan K, Çelik Ö, Bilgir E, Sağlam H, Kaplan FA, Görür SA, Odabaş A, Aslan AF, Różyło-Kalinowska I. Research Article A U-Net Approach to Apical Lesion Segmentation on Panoramic Radiographs. Biomed Res Int 2022;2022:7035367.
- 11. Çelik B, Savaştaer EF, Kaya HI, Çelik ME. The role of deep learning for periapical lesion detection on panoramic radiographs. Dentomaxillofac Radiol 2023;52:20230118.
- 12. Issa J, Jaber M, Rifai I, Mozdziak P, Kempisty B, Dyszkiewicz-Konwińska M. Diagnostic Test Accuracy of Artificial Intelligence in Detecting Periapical Periodontitis on Two-Dimensional Radiographs: A Retrospective Study and Literature Review. Medicina 2023;59:768.
- 13. Zadrożny Ł, Regulski P, Brus-Sawczuk K, Czajkowska M, Parkanyi L, Ganz S, Mijiritsky E. Artificial Intelligence Application in Assessment of Panoramic Radiographs. Diagnostics 2022;12:224.
Abstract
Objective: This study aimed to compare the performance of two deep learning models, YOLOv8 and Cascade Mask R-CNN, for the artificial intelligence-assisted automatic detection of periapical lesions in dental panoramic radiographs.
Material and Methods: A total of 1861 panoramic radiographs were reviewed, and 602 images containing periapical lesions were included in the study. In these radiographs, a total of 1059 lesions were identified. Observer agreement for lesion detection was evaluated using Kappa analysis. The included radiographs were divided into three datasets: training (70%), validation (20%), and testing (10%). The YOLOv8 and Cascade Mask R-CNN models were trained using the PyTorch library. The trained models were then applied to the test dataset, and true positive, false positive, true negative, and false negative outcomes were recorded for periapical lesion detection. Based on these outcomes, accuracy, sensitivity, specificity, precision, and F1 score were calculated for each model.
Results: On the test dataset, the YOLOv8 model achieved 96.9% accuracy, 53.5% sensitivity, 90% precision, and an F1 score of 0.671, whereas the Cascade Mask R-CNN model achieved 97.6% accuracy, 76.2% sensitivity, 81.9% precision, and an F1 score of 0.790. Although YOLOv8 produced fewer false positives, it failed to detect a considerable number of lesions. In contrast, Cascade Mask R-CNN was able to detect more lesions but generated relatively more false positives.
Conclusion: Artificial intelligence models demonstrated potential as supportive tools in clinical decision support systems based on their accuracy rates. However, the current sensitivity levels were insufficient to reliably undertake diagnostic responsibilities alone, suggesting that these models should not yet be used independently for clinical decision-making.
Ethical Statement
This study was conducted with the approval of the Ethics Committee of Karamanoğlu Mehmetbey University Faculty of Medicine (Approval No: 09-2024/01)
Supporting Institution
This study was supported by the KOSGEB (Approval no: 7AHNN).
References
- 1. Figdor D. Apical periodontitis: a very prevalent problem. Oral Surg Oral Med Oral Pathol 2002;94:651-652.
- 2. Bender I, Seltzer S. Roentgenographic and direct observation of experimental lesions in bone: I. J Am Dent Assoc 1961;62:152-160.
- 3. Bender I, Seltzer S. Roentgenographic and direct observation of experimental lesions in bone: II. J Am Dent Assoc 1961;62:708-716.
- 4. Jorge EG, Tanomaru-Filho M, Gonçalves M, Tanomaru JM. Detection of periapical lesion development by conventional radiography or computed tomography. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2008;106:56-61.
- 5. Dutra KL, Haas L, Porporatti AL, Flores-Mir C, Santos JN, Mezzomo LA, Correa M, Canto GDL. Diagnostic accuracy of cone-beam computed tomography and conventional radiography on apical periodontitis: a systematic review and meta-analysis. J Endod 2016;42:356-364.
- 6. Stera G, Giusti M, Magnini A, Calistri L, Izzetti R, Nardi C. Diagnostic accuracy of periapical radiography and panoramic radiography in the detection of apical periodontitis: a systematic review and meta-analysis. Radiol Med 2024;129:1682-1695.
- 7. Szabó V, Orhan K, Dobó-Nagy C, Veres DS, Manulis D, Ezhov M, Sanders A, Szabó BT. Deep Learning-Based Periapical Lesion Detection on Panoramic Radiographs. Diagnostics 2025;15:510.
- 8. Ba-Hattab R, Barhom N, Osman SAA, Naceur I, Odeh A, Asad A, Al-Najdi SARN, Ameri E, Daer A, Silva RLBD, Costa C, Cortes ARG, Tamimi F. Detection of Periapical Lesions on Panoramic Radiographs Using Deep Learning. Appl Sci 2023;13:1516.
- 9. Landis JR, Koch GG. The Measurement of Observer Agreement for Categorical Data. Biometrics 1977;33:159-174.
- 10. Bayrakdar IS, Orhan K, Çelik Ö, Bilgir E, Sağlam H, Kaplan FA, Görür SA, Odabaş A, Aslan AF, Różyło-Kalinowska I. Research Article A U-Net Approach to Apical Lesion Segmentation on Panoramic Radiographs. Biomed Res Int 2022;2022:7035367.
- 11. Çelik B, Savaştaer EF, Kaya HI, Çelik ME. The role of deep learning for periapical lesion detection on panoramic radiographs. Dentomaxillofac Radiol 2023;52:20230118.
- 12. Issa J, Jaber M, Rifai I, Mozdziak P, Kempisty B, Dyszkiewicz-Konwińska M. Diagnostic Test Accuracy of Artificial Intelligence in Detecting Periapical Periodontitis on Two-Dimensional Radiographs: A Retrospective Study and Literature Review. Medicina 2023;59:768.
- 13. Zadrożny Ł, Regulski P, Brus-Sawczuk K, Czajkowska M, Parkanyi L, Ganz S, Mijiritsky E. Artificial Intelligence Application in Assessment of Panoramic Radiographs. Diagnostics 2022;12:224.
Details
| Primary Language | English |
|---|---|
| Subjects | Oral and Maxillofacial Radiology, Endodontics |
| Journal Section | Research Article |
| Authors | |
| Submission Date | May 28, 2025 |
| Acceptance Date | July 11, 2025 |
| Publication Date | December 29, 2025 |
| Published in Issue | Year 2025 Volume: 28 Issue: 4 |
Cumhuriyet Dental Journal (Cumhuriyet Dent J, CDJ) is the official publication of Cumhuriyet University Faculty of Dentistry. CDJ is an international journal dedicated to the latest advancement of dentistry. The aim of this journal is to provide a platform for scientists and academicians all over the world to promote, share, and discuss various new issues and developments in different areas of dentistry. First issue of the Journal of Cumhuriyet University Faculty of Dentistry was published in 1998. In 2010, journal's name was changed as Cumhuriyet Dental Journal. Journal’s publication language is English.
CDJ accepts articles in English. Submitting a paper to CDJ is free of charges. In addition, CDJ has not have article processing charges.
Frequency: Four times a year (March, June, September, and December)
IMPORTANT NOTICE
All users of Cumhuriyet Dental Journal should visit to their user's home page through the "https://dergipark.org.tr/tr/user" " or "https://dergipark.org.tr/en/user" links to update their incomplete information shown in blue or yellow warnings and update their e-mail addresses and information to the DergiPark system. Otherwise, the e-mails from the journal will not be seen or fall into the SPAM folder. Please fill in all missing part in the relevant field.
Please visit journal's AUTHOR GUIDELINE to see revised policy and submission rules to be held since 2020.