A feasibility study for predicting optimal radiation therapy dose distributions of prostate cancer patients from patient anatomy using deep learning.

摘要: With the advancement of treatment modalities in radiation therapy for cancer patients, outcomes have improved, but at cost increased plan complexity and planning time. The accurate prediction dose distributions would alleviate this issue by guiding clinical optimization to save time maintain high quality plans. We modified a convolutional deep network model, U-net (originally designed segmentation purposes), predicting from patient image contours target volume (PTV) organs risk (OAR). show that, as an example, we are able accurately predict intensity-modulated (IMRT) prostate where average Dice similarity coefficient is 0.91 when comparing predicted vs. true isodose volumes between 0% 100% prescription dose. value absolute differences [max, mean] found be under 5% dose, specifically each structure [1.80%, 1.03%](PTV), [1.94%, 4.22%](Bladder), 0.48%](Body), [3.87%, 1.79%](L Femoral Head), [5.07%, 2.55%](R [1.26%, 1.62%](Rectum) thus managed map desired distribution patient's PTV OAR contours. As additional advantage, relatively little data was used techniques models described paper.