Abstract: Objective To establish a bone knife deformation prediction model for periacetabular osteotomy and quickly and accurately predict bone knife deformation. Methods A finite element numerical model of a pelvic bone knife containing both cortical and cancellous bones was established, and the correlation between nodal strain and deformation was analyzed. The strains of 5 nodes with the strongest integrated correlation were selected as the inputs, and the displacement increments of the nodes on the blade part were established as the outputs. After training the model with the dataset, a deep learning neural network regression model based on the finite element dataset was used to establish a prediction model for the strain deformation of the bone knife. After the model prediction was completed, a binocular visual localization system was used to determine the spatially accurate position of the bone knife during the osteotomy procedure as a means of intraoperative tracking of the bone knife. Results The R2 value of the prediction model was 0.987 81 and the average deformation error after discretizing the bone knife into nodes was 0.07 mm. The prediction model quickly and accurately acquired bone knife deformation and showed great potential for reducing PAO surgical accidents. Conclusions The bone knife deformation prediction model developed in this study rapidly predicted bone knife deformation from strain information. Thus, it can avoid injuring tissues, such as nerves and blood vessels around the tissue, reduce the difficulty and risk of periacetabular osteotomy, and provide theoretical support for clinical application.