We are pleased to announce that our latest research has been published in Oral Surgery, Oral Medicine, Oral Pathology and Oral Radiology (Elsevier).
The study, titled “Automated Detection of Patient Positioning Errors Using Mandibular Region-Focused Deep Learning Analysis,” develops a deep learning system that detects and classifies six types of patient positioning errors in dental panoramic radiography. On 1,497 radiographs (691 error-free, 806 erroneous), YOLOv10 and Mask R-CNN were compared for mandibular region extraction, and nine CNNs, three Vision Transformers and a novel ConvNeXt–Swin hybrid were evaluated on full images versus isolated mandibular regions with 5-fold cross-validation.
YOLOv10 reached an mAP@0.5 of 0.976 for mandibular extraction, and the hybrid model trained on mandibular images gave the best results in both binary (73.02% accuracy, F1 0.7280) and six-class (56.91%) classification. The study shows that region-focused analysis and CNN–Transformer hybrids outperform full-image and single-architecture approaches, offering a route to standardized quality control in clinical radiography workflows.
Authors:
Yasin Yaşa (Ordu University / Case Western Reserve University)
Hamza Osman İlhan (Yıldız Technical University)
Şevval Tuğçe Badik (Yıldız Technical University)
Gamzenur Güneş (Ordu University)
Furkan Özbey (Afyonkarahisar Health Sciences University)
Ali Zakr Syed (Case Western Reserve University)
