Options
Application of artificial intelligence and 3D printing in surgical planning for transforaminal lumbar interbody fusion
Type
thesis
Date Issued
2024-06-20
Author(s)
BUI TUAN ANH
Advisor
吳孟晃; 黃聰仁
Subjects
系所名稱：國際醫學研究博士學位學程
Publisher
國際醫學研究博士學位學程
Description
學位別：博士
關鍵字：Aritifical intelligence; 3D printing; Fusion surgery; Intervertebral cage
論文公開日期：2026-12-18
關鍵字：Aritifical intelligence; 3D printing; Fusion surgery; Intervertebral cage
論文公開日期：2026-12-18
Abstract
Background: Transforaminal lumbar interbody fusion (TLIF) is crucial for managing lumbar degenerative diseases, requiring precise cage height selection and lumbar lordosis correction. Traditional methods rely on subjective judgment, while emerging technologies like 3D printing and artificial intelligence (AI) offer promising solutions. This study evaluated the practical impact of AI technology and the potential of 3D printing in TLIF surgeries through comparative analysis.
Methods: The AI-driven process comprised of two key phases. Initially, a deep learning framework was employed to extract essential characteristics from X-ray images, which were then integrated with clinical data to formulate machine learning (ML) models. Subsequently, five ML algorithms underwent training to determine the most effective models for predicting interbody cage height and postoperative PI-LL. Following this, a subset of significant features from each baseline model was selected, and further analysis was conducted to examine the most critical features. As for the 3D-printing model, 20 cases were selected to construct the 3D model, which served as an external validation for the AI model.
Results: The AI model internally predicted cage height with an RMSE of 1.01, accurately predicting 131 out of 311 cases (42.12%), with 1 mm errors in the remaining cases. For PI-LL prediction, it achieved an RMSE of 5.19 and an accuracy of 0.81 for PI-LL stratification. In external validation, the model correctly predicted cage height in 11 out of 20 cases (55%). RMSE and MAE for PI-LL prediction were 3.28 and 2.91, respectively, versus 5.19 and 3.86 in internal validation. In constructing the 3D-printing model, it yielded RMSE and MAE values of 0.59 and 0.25, respectively, with an accuracy of 75%. For the prediction of PI-LL, the 3D model demonstrated lower RMSE and MAE values (2.62 and 2.02) than AI model. While there was no significant difference between AI and 3D printing in cage height prediction (p-value = 0.249), the models differed significantly in PI-LL prediction (p-value = 0.037).
Conclusion: Our study demonstrated the significant potential of AI and 3D printing in enhancing the accuracy of TLIF surgeries.
Methods: The AI-driven process comprised of two key phases. Initially, a deep learning framework was employed to extract essential characteristics from X-ray images, which were then integrated with clinical data to formulate machine learning (ML) models. Subsequently, five ML algorithms underwent training to determine the most effective models for predicting interbody cage height and postoperative PI-LL. Following this, a subset of significant features from each baseline model was selected, and further analysis was conducted to examine the most critical features. As for the 3D-printing model, 20 cases were selected to construct the 3D model, which served as an external validation for the AI model.
Results: The AI model internally predicted cage height with an RMSE of 1.01, accurately predicting 131 out of 311 cases (42.12%), with 1 mm errors in the remaining cases. For PI-LL prediction, it achieved an RMSE of 5.19 and an accuracy of 0.81 for PI-LL stratification. In external validation, the model correctly predicted cage height in 11 out of 20 cases (55%). RMSE and MAE for PI-LL prediction were 3.28 and 2.91, respectively, versus 5.19 and 3.86 in internal validation. In constructing the 3D-printing model, it yielded RMSE and MAE values of 0.59 and 0.25, respectively, with an accuracy of 75%. For the prediction of PI-LL, the 3D model demonstrated lower RMSE and MAE values (2.62 and 2.02) than AI model. While there was no significant difference between AI and 3D printing in cage height prediction (p-value = 0.249), the models differed significantly in PI-LL prediction (p-value = 0.037).
Conclusion: Our study demonstrated the significant potential of AI and 3D printing in enhancing the accuracy of TLIF surgeries.