Surgeons routinely interpret preoperative radiographic images for estimating the shape and position of the tooth prior to performing tooth extraction. A recent study aimed to predict the difficulty of lower wisdom tooth extraction using only panoramic radiographs, as surgeons often encounter complications and difficulty when dealing with these extractions.
The study evaluated the performance of deep learning models—AlexNet and VGG-16—to predict the necessity for tooth separation or bone removal during wisdom tooth extraction. Both surgeons and the deep learning models interpreted the same set of images to classify extraction difficulty based on the modified Parant score.
The results indicated the challenges of predicting the extraction difficulties using only panoramic radiographs, with human accuracy averaging 54.4% for both oral surgeons, compared to 57.7% for AlexNet and 54.4% for VGG-16. Despite these limitations, both deep learning models showcased high sensitivity, exceeding 90% for crown and root separation, demonstrating comparable predictive ability to experienced oral surgeons.
Extraction of impacted mandibular wisdom teeth often results in complications such as bleeding, nerve damage, and infections, making it imperative for oral surgeons to have accurate preoperative predictions. Yet, factors like tooth inclination, proximity to adjacent structures, and root morphology complicate these predictions, leading to reliance on traditional methods like panoramic radiographs.
Although many classifications exist for estimating extraction difficulty, this study points to their limitations, stating, "Accurately predicting the difficulty of wisdom teeth extraction using panoramic radiographs alone is challenging." The study leads to the conclusion advocating for the adoption of advanced imaging modalities such as cone-beam computed tomography (CBCT) or supplementary diagnostic tools to achieve higher precision in surgical predictive capabilities.
Deep learning shows promise as it could revolutionize the predictive process. Both AlexNet and VGG-16 models demonstrated sensitivities over 90%, particularly for complex procedures, indicating they are effective as screening methods.
Before deep learning models can replace traditional methods, validation through multicenter studies with broader sample sizes is required. This is the first study to develop deep learning models capable of predicting the extraction difficulty using actual surgical results. The findings highlight the limitations of panoramic radiographs and the need for integrating advanced imaging techniques to improve preoperative assessments.
The study concludes by emphasizing the necessity for advancements in diagnostic tools and techniques, stating, "The predictive ability ... is equivalent to ... of an oral surgery specialist, and ... makes it suitable for screening in clinical settings," showcasing the relevance of deep learning applications to benefit patients and surgeons alike.
