A novel AI model for detecting periapical lesion on CBCT: CBCT-SAM

Area of Science:

• Dentistry
• Radiology
• Artificial Intelligence

Background:

• Periapical lesions can be subtle and may be missed on radiographic scans, particularly by less experienced dentists.
• Asymptomatic or early-stage lesions on cone-beam computed tomography (CBCT) require careful assessment, especially when scans are not primarily for endodontic evaluation.
• Existing algorithms aid radiographic assessment, but novel AI solutions are needed for improved diagnostic accuracy.

Purpose of the Study:

• To evaluate the efficacy of a new artificial intelligence (AI) model, CBCT-SAM, in detecting periapical lesions on CBCT images.
• To compare the diagnostic and segmentation performance of CBCT-SAM against other AI models and assess its potential for clinical application.

Main Methods:

• CBCT scans from 185 patients with confirmed periapical lesions were used for model training and validation.
• Manual segmentation was performed by a trained operator and validated by a maxillofacial radiologist.
• The diagnostic and segmentation performance of CBCT-SAM (with and without its progressive Prediction Refinement Module) was compared to Modified U-Net and PAL-Net using metrics like accuracy, sensitivity, specificity, precision, and Dice Similarity Coefficient (DSC).

Main Results:

• CBCT-SAM achieved high diagnostic accuracy (98.92% ± 10.37%) and segmentation accuracy (99.65% ± 0.66%).
• CBCT-SAM and PAL-Net significantly outperformed Modified U-Net in segmentation accuracy, sensitivity, and DSC.
• The inclusion of the progressive Prediction Refinement Module in CBCT-SAM led to a slight improvement in diagnostic and segmentation performance compared to PAL-Net.

Conclusions:

• CBCT-SAM demonstrates expert-level capability in identifying periapical lesions on CBCT scans.
• The integration of AI in radiographic assessment can enhance dentists' diagnostic accuracy and efficiency, reducing missed diagnoses and facilitating early treatment.