MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

A Method to Predict the Patient-Specific Dose-Volume Histogram Curves for Radiotherapy Planning with Deep Learning

X Chen*, K Men, J Yi, J Dai, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences, Beijing, 100021,China

Presentations

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

Room: AAPM ePoster Library

Purpose: The automatic treatment planning tools need to set the reverse optimization parameters based on the desired Dose-Volume Histogram (DVH). Current deep learning methods could predict the three-dimensional dose distribution and then calculate the DVH curves manually. In this study, we developed an end-to-end method to predict patient-specific DVH directly for radiotherapy planning with deep learning.

Methods: Patients data included 80 cases with nasopharyngeal cancer, of which 70 cases for training and 10 cases for testing. A convolutional neural networks (CNN) model was trained to perform prediction of the DVH. Structure images (StrutImgs) with contoured structures corresponding to CT images were generated as the inputs, which reflected the spatial position information of organs at risk (OARs) and targets. The DVH curves with 256 nodes were the outputs. The accuracy of prediction was evaluated against the corresponding DVH generated from expert plans. Six OARs were involved in the study for DVH prediction and a global DVH analysis was calculated for evaluation.

Results: The proposed method can generate the DVH curve accurately. The average relative MEDVH values for brain stem PRV,left mandible,right mandible,left parotid,right parotid and spinal cord PRV were 0.68%,0.92%,0.94%,-4.10%,2.11% and -1.65%. The whole predicted DVH of left parotid was lower than the ground truth value which had the significant statistic difference.

Conclusion: The proposed method could automatically generate accurate patient-specific DVH for radiotherapy. It can be applied in planning quality assurance and automated planning. One advantage of this method is that any type of CNN can be adapted for this task. The model can get better continuously with the deep learning algorithms improve.

Funding Support, Disclosures, and Conflict of Interest: This work was supported by the Beijing Hope Run Special Fund of Cancer Foundation of China (LC2018A14, LC2019B06), the Beijing Municipal Science & Technology Commission (Z181100001918002), and the National Natural Science Foundation of China (11975313).

Keywords

Dose Volume Histograms, Dosimetry, Inverse Planning

Taxonomy

TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation

Contact Email