MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Accuracy of a Novel Facial Feature Point Algorithm for Head Motion Tracking Using Surface Guided Imaging to Remove the Mask in Head and Neck Radiotherapy

Y Ben Bouchta1*, C Cheng1, K Makhija1, J Sykes2, E Steiner1, P Keall1 (1) ARCF Image-X Institute - The University of Sydney, Eveleigh, NSW, AU (2) Blacktown Hospital, Blacktown, NSW, AU

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose:
For up to 50% of head and neck(H&N) patients, the immobilisation mask can induce claustrophobia and anxiety leading in 24% of cases to treatment disruption and/or refusal. Alternatives to immobilisation masks require real-time tracking of patient motion in order to preserve accuracy. This work presents, to our knowledge, the first method of tracking head motion for surface guided radiotherapy (SGRT) using a facial feature points (FFP) algorithm.

Methods:
A dual depth camera system was used to capture a 3-dimensional model of the head of three volunteers asked to lay still for 5 minutes. The model was used to obtain a 2-dimensional image from which 68 FFP were extracted and back-projected on the 3-dimensional model. The noise on individual FFP was measured and the best 17 FFP were kept for head motion tracking.
The accuracy of head motion measurements was estimated in silico by comparing two distinct, but registered images. One image was used as a reference while the second image was transformed using a randomly generated 6 degree-of-freedom transform. The inverse transform was used as the ground truth. The transformed image was matched with the reference image using the FFP method and the difference between the two was recorded. This process was repeated 22500 times.


Results:
The accuracy of our localisation method differed for each FFP but was consistent across the 3 volunteers. The mean error for the 17 FFP was of 0.8 mm ranging from 0.6 to 0.9 mm. The mean error in head motion was 1±0.7 mm, 0.6±0.4 mm and 0.9±0.7 mm in the left-right, inferior-superior and anterior-posterior axes respectively. Rotational uncertainty was < 0.5 degree.

Conclusion:
FFP-based algorithms appear to be a promising way of tracking head motion for H&N SGRT. However, further developments are needed to improve the accuracy of head motion tracking.

Funding Support, Disclosures, and Conflict of Interest: The authors would like to acknowledge the Cancer Institute of New South Wales Translational Program Grant for funding this project.

Keywords

Surface Matching

Taxonomy

Not Applicable / None Entered.

Contact Email