Intelligent Prediction Of Orthodontic Treatment For Protrusive Facial Deformity

Develop a high-precision model for predicting soft/hard tissue changes in protrusive facial deformity after orthodontic treatment. Pretreatment and posttreatment lateral cephalograms of 1,052 patients with skeletal Class I and Class II protrusive deformities were collected. After standardized preprocessing, the preprocessed data was divided into training, validation, and test sets (7.5:0.5:2 ratio). The proposed P-CDGAN model combined a U-Net generator with PatchGAN discriminator. Model performance was evaluated using peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) . The mean absolute error (MAE) and successful detection rate (SDR) of 11 soft tissue and 6 hard tissue landmarks were calculated. Independent samples t-tests were used to compare landmark errors between groups. For skeletal Class II cases, the average MAE for soft/hard tissue landmarks was 2.03 ± 0.43 mm and 1.67 ± 0.23 mm, respectively, no significant intergroup differences were observed (P>0.05). Within a 2.0 mm error tolerance, the SDR for skeletal Class I soft/hard tissues reached 93.3%/94.1%, while skeletal Class II achieved 89.7%/91.4%. The predicted images exhibited an average PSNR of 23.69 and SSIM of 0.877. The P-CDGAN model is the first to achieve precise joint prediction of soft and hard tissue changes in protrusive facial deformity post-orthodontic treatment. The model demonstrates outstanding performance in complex skeletal Class II cases, highlighting its significant clinical value.