MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Artificial Intelligence-Based Dose-Guided Patient Positioning for Prostate Cancer Online Adaptive Radiotherapy

X Zhang1*, (1) West China hospital of Sichuan university, Chengdu, Sichuan

Presentations

(Tuesday, 7/16/2019) 1:15 PM - 1:45 PM

Room: Exhibit Hall | Forum 5

Purpose: Validation of planning CT to CBCT deformable image registration (DIR)-based delivered dose reconstruction, support vector regression-based dose interpolation for execute time reduction and conventional neural network-based dose-volume histograms (DVH) selection for optimal localization determination, corporately carrying forward an artificial intelligence-based dose-guided patient positioning for prostate cancer online adaptive radiotherapy.

Methods: Artificial intelligence-based dose-guided patient positioning is dosimetrically aligning planning CT to CBCT without human intervention. To execute dose-guided patient positioning in 15 mins, Fusion CT and deformable vector field-based OARs derived from planning CT to CBCT DIR were used for dose reconstruction. Feasibility of DIR-based dose calculation has been validated in our previous work. Based on IGRT alignment isocenter, 27 points of 4x4x4 mm3 cube with 2mm stepwise was selected as potential treatment isocenter. DVH parameter of nine cube vertexes was acquired using treatment planning system and DVH parameter of another eighteen points was predicted by means of support vector regression-based dose interpolation. Based on decision making of radiology oncologist, training conventional neural network to determine optimal localization. Besides, interchangeability of DVF-based OARs and manually delineate OARs was evaluated in terms of their corresponding DVH selection results. The workflow was retrospectively evaluated for 30 prostate patients with a total of 326 CBCTs.

Results: The accuracy of support vector regression-based dose interpolation and conventional neural network was found sufficient, facilitating the implementation of artificial intelligence-based dose-guided patient positioning. Compared to IGRT, the mean prescription dose coverage to primary lesion could be improved by 4.3% while OARs within clinical tolerance limitation. Difference of DVF-based OARs and manually delineate OARs has minor impact on DVH selection.

Conclusion: Artificial intelligence-based dose-guided patient positioning is feasible and offers increased control over target coverage.

Keywords

Patient Positioning, Radiation Therapy, Dose Volume Histograms

Taxonomy

IM- Cone Beam CT: Machine learning, computer vision

Contact Email