MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All
  4. Applications of Deep Learning in Patient-Specific QA of MLC-Equipped Robotic SBRT and SRS

Applications of Deep Learning in Patient-Specific QA of MLC-Equipped Robotic SBRT and SRS

B Han*, L Xing , L Wang , Stanford University School of Medicine, Stanford, CA

Presentations

(Sunday, 7/14/2019)  

Room: ePoster Forums

Purpose: To systematically investigate the applications of state-of-the-art deep learning techniques for accurate patient-specific quality assurance (QA) of MLC-based SRS and SBRT treatments.

Methods: The commissioning and clinical measurement data were collected from our institution and the datasets. A total of 200 MLC-formed fields with standard square and rectangular shapes and irregular shape from clinical plans were analyzed. 80% and 20% of the data were randomly split into training and testing subsets. A deep learning model was created using the Tensorflow packages in python. The model was consisted of 4 convolution layers to extract features and two fully connected layers for the final predictions. The measured dose distributions were treated as outputs with inputs of different fields with detailed MLC positions. Dose results for different MLC fields were predicted using models trained with regularization added to the cost functions to prevent overfitting. The predicted dose distributions for small and irregular fields were evaluated using percentage relative error regarding measured data at the depth of 1.5cm and 5cm.

Results: With augmentation techniques, datasets of regular and irregular shaped fields with sizes ranging from 7mm x 7mm up to 115mm x 100mm were tested for model training, testing and dose output prediction. The dose of small and irregular SRS and SBRT treatment field was accurately predicted with the proposed deep learning methods.  The mean relative error between the predicted and the measured dose is 0.11% with a maximum error of 0.4%. The prediction of off-axis dose distribution was also tested with a good agreement within 1% compared to the film and ion chamber measurement data. 

Conclusion: The proposed deep learning based method was validated as an accurate dose verification tool and efficient patient-specific QA tool for MLC-based Robotic SRS and SBRT.

Keywords

Stereotactic Radiosurgery, Quality Assurance, Dosimetry

Taxonomy

Not Applicable / None Entered.

Contact Email

Additional Links