Predicting occlusal force distribution from intraoral scans using deep learning: a pilot study

Clinical Snapshot

Key Findings

• P Value: Overall model comparison: p=0.00069 (Friedman test). Post-hoc Holm-corrected Wilcoxon: CNN-Attention vs U-Net, Vision Transformer, and ResNet-UNet adjusted p<0.05; CNN-Attention vs Pix2Pix: not significant

• Effect Size: CNN-Attention achieved the highest SSIM (mean 0.501 ± 0.030); Pix2Pix achieved the lowest MAE (0.237 ± 0.018). Statistically significant inter-model difference: Friedman χ²=19.3

• Primary Outcome: Structural Similarity Index (SSIM) between deep learning-predicted and ground-truth T-Scan occlusal force maps across five architectures

• Nnt Or Sensitivity: No sensitivity, specificity, AUC, or NNT reported. SSIM of 0.501 (scale 0–1) indicates moderate structural similarity; no established clinical threshold for this application exists

• Confidence Interval: Bootstrap confidence intervals were calculated but specific values are not reported in the abstract

Clinical Application

Conceptually feasible within existing digital dentistry infrastructure — intraoral scanners are widely available in specialist and general dental practices. However, the technology requires substantial further development, including large-scale dataset acquisition, prospective clinical validation, regulatory approval, and integration into commercial CAD/CAM or dental practice management software before any clinical deployment is feasible. Intraoral scanning is increasingly adopted in Australian specialist prosthodontic and orthodontic practices, and is supported within RACGP-adjacent dental guidelines for digital workflows. T-Scan occlusal analysis systems are available in Australia but not universally used in general practice due to cost. The TGA would require prospective clinical validation data before any AI-based occlusal prediction tool could be registered as a medical device (Class IIa or higher under the TGA Medical Devices framework). PBS does not currently fund digital occlusal analysis. Australian dental schools and university clinics (e.g., University of Melbourne, University of Sydney, Griffith) represent appropriate settings for future multicentre validation studies. The RACGP has no specific guidelines on AI-assisted occlusal analysis, reflecting the early-stage nature of this technology. Not yet applicable to any clinical population. At this stage, findings are relevant only as a proof-of-concept for researchers in digital dentistry and dental informatics. Future applicability may extend to patients undergoing full-arch prosthetic rehabilitation, implant planning, or occlusal splint fabrication where functional force data would complement morphological scan data.

Abstract

OBJECTIVES: This pilot study investigated whether occlusal force distribution measured by T-Scan can be predicted from intraoral scan images using deep learning. METHODS: Paired intraoral scan and T-Scan images from 26 subjects were analyzed. Following quality control, data were randomly partitioned at the subject level into training (n = 18), validation (n = 4), and independent test (n = 4; 8 image pairs) sets. Five deep learning architectures (U-Net, ResNet-UNet, Vision Transformer, CNN-Attention, and Pix2Pix) were trained to predict T-Scan force maps from occlusal scan images. Structural Similarity Index (SSIM) was predefined as the primary endpoint. Nonparametric statistical comparisons (Friedman test with Holm-corrected Wilcoxon post hoc analysis) and bootstrap confidence intervals were used. RESULTS: A significant difference in SSIM was observed among models (χ² = 19.3, p = 0.00069). CNN-Attention achieved the highest structural similarity (mean SSIM 0.501 ± 0.030), significantly outperforming U-Net, Vision Transformer, and ResNet-UNet (adjusted p < 0.05), while not differing significantly from Pix2Pix. Pix2Pix demonstrated the lowest mean absolute error (0.237 ± 0.018), indicating superior pixel-level intensity approximation. Validation curves revealed distinct optimization behaviors, with adversarial training showing late-stage improvements in SSIM. CONCLUSIONS: Deep learning models approximated T-Scan occlusal force maps from intraoral scan images in this pilot dataset. These findings suggest that artificial intelligence may help estimate functional occlusal force distribution from morphological intraoral scan data. Further studies with larger datasets are required to validate clinical applicability. CLINICAL SIGNIFICANCE: Computational prediction of functional occlusal force maps from intraoral scans may represent a future step toward integrating morphological and functional data within digital dentistry workflows; however, larger validation studies are required before clinical application.

References

1. 1.Şahin, C. H., Bostancıoğlu, S. E., Ediz, İ., Toğay, A., & Tamam, E. (2026). Predicting occlusal force distribution from intraoral scans using deep learning: a pilot study. Clinical Oral Investigations. https://doi.org/10.1007/s00784-021-04191-1