Artifical Intelligence in Predicting Surgical Problems and Postoperative Morbidity in Mandibular Third Molar Extractions

Comprehensive Summary

This cross-sectional study by Karadiguddi et al. examined how effective AI algorithms were at predicting surgical difficulty and postoperative complications of mandibular third molar (MTM) extractions. Forty patients were recruited for this study and inclusion criteria required patients to have indication for MTM extraction and availability of preoperative cone-beam computed tomography (CBCT) scans. Exclusion criteria included pregnant women, those with systemic conditions affecting wound healing, and those with a history of MTM surgery. The primary outcomes included surgical difficulty (classified intraoperatively as easy, moderate, or difficult based on Pell and Gregory classification), operative time, need for sectioning, and postoperative morbidity which was assessed at 7 to 14 days postoperatively and encompassed pain, swelling, and nerve paresthesia. The authors established two AI models, Random Forest (RF) and Convolutional Neural Networks (CNNs), to predict these outcomes and assess their performance by analyzing their accuracy, sensitivity, specificity, and area under the curve. The RF model used CBCT and patient factors to predict surgical difficulty and the CNN model used the CBCT images to estimate the likelihood of postoperative morbidity. Overall, the RF model achieved an overall accuracy of 87.5% and an area under the receiver operating curve of 0.92, indicating very good predictive performance for surgical difficulty. The CNN model predicted postoperative morbidity with an accuracy of 82.3% and had an area under the curve of 0.88, showing a very good capacity in distinguishing likelihood for patients to develop postoperative complications. It was also determined that proximity of MTM roots to the inferior alveolar nerve within 2 mm raised odds of surgical difficulty and post-operative morbidity and patients over 30 years old had higher likelihood of adverse outcomes. Overall, AI models were effective at predicting these primary outcome variables.

Outcomes and Implications

The findings from this study suggest that AI models like Random Forest and CNNs can be highly reliable tools for planning dental surgeries. By accurately predicting surgical difficulty and potential complications, these algorithms help surgeons better prepare for "difficult" cases before the first incision is made. Using AI in this way could lead to shorter operative times and fewer post-operative issues like pain or swelling. Ultimately, integrating these models into clinical practice could improve patient safety and help dentists and potentially other healthcare providers offer more predictable surgical outcomes.