# A data-driven method for surgeon-specific difficulty assessment in third molar extraction

## TL;DR

This study introduces a data-driven method to assess the difficulty of wisdom tooth extraction for surgeons at different experience levels, improving training and evaluation.

## Contribution

A novel data-driven decoupling-prediction model for surgeon-specific difficulty assessment in third molar extraction is proposed.

## Key findings

- The proposed method achieved 80% accuracy and an AUC of 0.85 using SVM.
- Inexperienced surgeons are influenced by more factors than experienced ones, who are mainly affected by four key factors.
- Surgeons typically become proficient after 8 months of practice, as shown by learning curves.

## Abstract

The purpose of this study is to use a data-driven method to analyze the time taken by junior doctors to extract lower wisdom teeth and the factors affecting the difficulty of the procedure. It aims to reveal the distribution characteristics of difficulty factors at different stages of development, establish a mathematical model for procedural difficulty, evaluate the effectiveness of the existing difficulty scale, and provide difficulty indicators for the extraction training of impacted teeth for young doctors at different stages.

We collected surgical records of 419 cases of lower impacted wisdom teeth extraction completed by 9 residents. The difficulty index was based on a scale with 14 primary indicators and 37 secondary indicators. We proposed a data-driven method for surgeon-specific difficulty assessment (DDSS) of third molar extraction surgery. When assessing the surgical difficulty for a surgeon, the DDSS uses a method based on Lasso regression to classify the doctor as either a junior doctor who has completed grade 1 training or a novice doctor. It then calls upon the corresponding pre-trained model to conduct targeted difficulty prediction and provide key difficulty factors.

Our method achieved an accuracy of 80% and an AUC of 0.85 with SVM. The methods we proposed outperformed the methods without decoupling. The clustering analysis revealed that inexperienced surgeons are affected by a larger number of factors, while experienced surgeons are primarily influenced by four key factors: Crown resistance, impacted type, mouth opening, and gender. Learning curves indicated that surgeons typically become proficient after 8 months of practice.

We propose a data-driven decoupling-prediction model, which improves the model’s performance in the task of assessing dental surgery difficulty. We also draw the learning curve of novice surgeons based on the data decoupling method we proposed. This provides a new perspective for surgical difficulty assessment and surgeon training, and offers a reliable conclusion.