Estimation of lung tumor position from multiple anatomical features on 4D-CT using multiple regression analysis

Area of Science:

• Radiotherapy
• Medical Imaging
• Computational Anatomy

Background:

• Accurate lung tumor localization is critical for effective stereotactic body radiotherapy (SBRT).
• Four-dimensional computed tomography (4D-CT) captures respiratory motion but requires precise internal target volume (ITV) definition.
• Traditional ITV definition can be time-consuming and may overestimate tumor margins.

Purpose of the Study:

• To estimate lung tumor position using single regression analysis (SRA) and multiple regression analysis (MRA) on 4D-CT data.
• To evaluate the impact of SRA and MRA on ITV for lung SBRT.
• To compare the accuracy of tumor position estimation using different regression approaches.

Main Methods:

• Utilized 4D-CT data from 12 lung cancer cases with >5 mm tumor motion.
• Measured 3D tumor position and anatomical features (lung volume, diaphragm, chest/abdominal walls).
• Estimated tumor position via SRA and MRA, calculating root-mean-square error (RMSE) and comparing ITVs to conventional methods.

Main Results:

• Multiple regression analysis (MRA) demonstrated higher accuracy (RMSE within 1.6 mm) compared to single regression analysis (SRA) (RMSE within 3.7 mm).
• Lung volume was the most influential anatomical feature in MRA.
• MRA and SRA approaches reduced ITV by an average of 38.3% and 31.9%, respectively, compared to conventional ITV.

Conclusions:

• MRA provides a more accurate estimation of lung tumor position than SRA.
• Both SRA and MRA significantly reduce ITV, potentially minimizing radiation dose to healthy tissues.
• These regression-based approaches offer a promising method for improving lung SBRT planning.