Click here to


Are you sure ?

Yes, do it No, cancel

Utilizing Quantitative Local Trajectory Method to Online Analyse Intrafraction Prostate Motion

Y Gao*, B Zhao , X Qi , X Gao , Peking University First Hospital, Beijing


(Sunday, 7/14/2019)  

Room: ePoster Forums

Purpose: There are various intrafracton prostate movement patterns, and they were mainly classified qualitatively about a whole fraction after treatment in previous studies. So we proposed a new method, called quantitative local trajectory (QLT) method, which can not only quantitatively analyze but online discriminate different pattern by segmenting motion trajectories.

Methods: 1268 intrafraction motion trajectories recorded by ultrasound real-time tracking system in 61 patients can be considered as four typical patterns: stable target at baseline, persistent excursion, transient excursion and continuous drift. Our QLT method can be implemented in four steps: firstly, a single motion trajectory was divided into ‘segments’ in order to obtain the above four typical patterns; secondly, useful quantitative factors were selected by stepwise discriminant analysis; thirdly, different patterns were identified online and accuracy of different data sets were calculated after discriminant functions were obtained; finally, clinical analysis of this algorithm was preliminary explored in terms of dosimetric deviation.

Results: 2353 ‘segments’ were obtained by segmentation. Eight quantitative factors were selected by stepwise discriminant analysis, and significant statistical difference was observed among four patterns (p<0.01). The discrimination accuracy of QLT method was 97.7% for the training set as well as 96.4% for the testing set. The target coverage deviation between QLT method patterned intrafraction motion trajectory and actual trajectory were all less than 1% for the typical ‘segments’.

Conclusion: Quantitative local trajectory method can be used to online analyze intrafraction prostate motion including effectively discriminate different movement patterns and accurately calculate different effects on dosimetry. Further, the individualized intervention thresholds can be optimized and previous or online quality control strategies can be implemented. In conclusion, this method has great potential in clinical application.


Not Applicable / None Entered.


IM- Dataset analysis/biomathematics: Machine learning

Contact Email