Author: Mingli Chen, Huan Amanda Liu, Weiguo Lu, Lin Ma 👨🔬
Affiliation: Medical Artificial Intelligence and Automation (MAIA) Lab, Department of Radiation Oncology, UT Southwestern Medical Center, Mayo Clinic 🌍
Purpose: To reduce the back-and-forth in planning process between physicians and dosimetrists resulting from trade-off exploration, we proposed a novel deep learning framework called DeepTuning.
Methods: DeepTuning can predict doses with different trade-offs by manipulating the deepest layer Z (a 1x1x1024 vector). DeepTuning leverages two encoders for prior and posterior inference. Prior encoder just takes contours as input and extracts geometric information for conventional dose prediction (predicting "average" dose with no trade-off). Posterior encoder takes both contour and dose as input and extracts ΔZ that encodes the trade-off of input dose. Given a template plan (treated patient) with desired trade-off, posterior inference can extract trade-off information ΔZ. When predicting dose for a new patient with just contours, prior inference (prior encoder+decoder) can predict not only an "average" dose, but also doses with desired tradeoffs when we apply the extracted ΔZs.
Results: We validated DeepTuning with a prostate dataset composed of 99 cases. We retrospectively optimized two VMAT plans for each case, prioritizing PTV coverage (pro-ptv) and rectum sparing (pro-oar), respectively. DeepTuning was trained, validated and tested by 60, 10 and 29 cases. The baseline is prior inference route that predicts fixed “average” dose distributions. The mean rectum doses are 49.5±9.0Gy, -3.1±5.3% cooler than ground truth (GT) pro-ptv doses (52.0±10.6Gy) and 6.6±7.9% hotter than GT pro-oar doses (44.2±12.2Gy). Then we extracted ΔZs for pro-ptv trade-off and pro-oar trade-off from the two plans of a training case. With ΔZs applied, DeepTuning can predict doses with two different trade-offs. The mean rectum doses of the pro-ptv predictions are 51.8±8.5Gy, -0.27±5.1% different from GT, while pro-oar predictions are 43.8±10.1Gy, -0.6±7.5% away from GT.
Conclusion: DeepTuning empowers physicians to tune doses and explore trade-offs immediately after contouring. As the trade-off exploration occurs before dosimetry planning, DeepTuning can reduce back-and-forth and potentially revolutionize treatment planning workflow.