Stacked ensembles accurately detect and stratify periapical lesion size on intraoral radiographs

Stacked deep learning ensembles achieved 0.96 accuracy and 1.00 macro-AUC for periapical lesion detection, with 0.94 sensitivity for lesions <2 mm. Small internal study; external validation needed.

• Retrospective study on 146 intraoral radiographs with cropped root-apex regions and augmentation.
• Stacked ensemble (multinomial logistic regression meta-learner) achieved accuracy 0.96 and macro-AUC 1.00, outperforming individual CNNs.
• Sensitivity for sub-2mm lesions was 0.94; single-centre internal test, requiring prospective multicentre external validation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.