Abstract: Background　Chronic apical periodontitis is a common oral infectious disease, and the precise automatic segmentation of its lesion areas is of great clinical significance for assisting treatment planning.Objective　A new type of network combining the SAM3 basic large model with the shift attention mechanism is proposed to identify and segment the chronic periapical periodontitis lesions on periapical radiographs.Methods　This study was a single-center retrospective study. Data from 500 cases in the Department of Oral Medicine of the First Medical Center of the Chinese PLA General Hospital were collected. Two oral radiology doctors marked the lesion areas, and one senior expert arbitrated the results. The Shift-SAM network was used to segment and identify the chronic apical periodontitis lesion areas on the periapical radiographs, and was compared with 12 commonly used deep learning models. An ablation experiment was also set up. The segmentation performance was evaluated using Dice coefficient, intersection-over-union ratio, sensitivity, specificity, F2 score, and Hausdorff distance.Results　The intersectionover- union ratio, sensitivity, precision, F2 score and Hausdorff distance of the Shift-SAM model reached 0.622, 0.484, 0.623, 0.697 and 0.617 respectively. Its overall performance is significantly superior to the existing U-Net, Mamba, Transformer and SAM-like comparison models, and it has certain potential for clinical application.Conclusion　The novel Shift-SAM network proposed in this study overcomes the shortcomings of the basic model in terms of local texture noise resistance through the shift attention mechanism, and is capable of identifying and segmenting the lesion areas of chronic periapical periodontitis on periapical radiographs.