Related Experiment Video
Updated: Jun 20, 2026
08:17
Dynamic Lung Tumor Tracking for Stereotactic Ablative Body Radiation Therapy
Published on: June 7, 2015
Predicting respiratory tumor motion with multi-dimensional adaptive filters and support vector regression
Nadeem Riaz1, Piyush Shanker, Rodney Wiersma
1Department of Radiation Oncology, Stanford University, 875 Blake Wilbur Drive, Stanford, CA 94305-5847, USA. nriaz@stanford.edu
Physics in Medicine and Biology
|September 5, 2009
Summary
Predicting lung tumor motion during radiotherapy is crucial for accurate treatment. Support vector regression offers the most accurate prediction, achieving less than 2 mm error at 1-second latency, improving radiation delivery.
Area of Science:
- Medical Physics
- Radiotherapy Technology
- Computational Biology
Background:
- Intra-fraction tumor motion tracking enhances radiation delivery accuracy.
- Real-time tumor tracking introduces latency, requiring future position prediction.
- Accurate tumor motion prediction is vital for effective radiotherapy gating and beam tracking.
Purpose of the Study:
- To evaluate multi-dimensional linear adaptive filters and support vector regression for lung tumor motion prediction.
- To compare novel prediction methods against conventional techniques like linear regression and single-input, single-output adaptive filters.
- To assess prediction accuracy at varying latencies (400 ms and 1 s).
Main Methods:
- Utilized a multi-dimensional linear adaptive filter framework (MISO) for tumor motion prediction.
- Applied Support Vector Regression (SVR) for predicting future tumor positions.
- Compared MISO and SVR against linear regression and single-output adaptive filters using 30 Hz tracked lung tumor data.
Main Results:
- Support Vector Regression (SVR) demonstrated the highest accuracy, with root-mean-square-errors (RMSEs) of 1.26 mm at 400 ms and 1.93 mm at 1 s latency.
- Multi-dimensional adaptive filters showed improved performance over single-dimension filters.
- Conventional methods like linear regression and single-output adaptive filters yielded higher RMSEs compared to SVR and MISO.
Conclusions:
- Support Vector Regression is a highly effective method for predicting lung tumor motion in radiotherapy.
- Multi-dimensional adaptive filtering frameworks enhance prediction accuracy.
- Further research aims to combine SVR and MISO frameworks for optimal tumor motion prediction and improved radiotherapy outcomes.
