AI Read Baseline Radiographs to Predict Peri-Implantitis With Moderate Accuracy

Edited by Dentaltown staff

Deep-learning models applied to routine baseline radiographs predicted which dental implants would later develop peri-implantitis with moderate accuracy, according to a multicenter retrospective study published online Oct. 19 in Clinical Oral Investigations.

The study included 195 implants in 108 maintained patients: 146 implants from the University of Michigan and 49 from King’s College London. For each implant, investigators analyzed a baseline periapical radiograph taken up to one year after prosthetic restoration. Patients had at least two years of documented follow-up after loading.

The team tested five convolutional neural network architectures pretrained on large image datasets: ResNet50, DenseNet, VGG16, InceptionV3, and EfficientNet. Each was combined with four transfer learning strategies. Models were trained using patient-level splits to avoid data leakage and evaluated on an independent test set.

Non-hybrid feature extraction produced the most consistent results, with the best models reaching areas under the curve of 0.74 to 0.83. InceptionV3 performed best, at 0.83. Fine-tuning strategies performed modestly or poorly, with AUC values generally between 0.42 and 0.69.

Muhammad H. A. Saleh and Hom-Lay Wang of the University of Michigan School of Dentistry, Luigi Nibali of King’s College London, and Giuseppe Troiano of LUM University in Italy conducted the study with colleagues at the University of Foggia and the University of Tripoli. It was partially supported by the Graduate Periodontics fund at the University of Michigan School of Dentistry, and the authors declared no competing interests.

The authors said transfer learning may help with early risk stratification but that larger prospective studies are needed before clinical adoption. They listed several limitations: the limited number of patients from two centers, the mix of implant systems and prosthetic designs, and reliance on radiographs without soft tissue variables.

Sources:
Clinical Oral Investigations, “Accuracy of transfer learning in predicting peri-implantitis onset from intraoral radiograph: A multi-center longitudinal retrospective study,” published online Oct. 19, 2026: link.springer.com/10.1007/s00784-026-07181-3