Purpose: We propose a treatment planning framework that accounts for weekly lung tumor shrinkage using CBCT images with a deep learning-based model.
Methods: Sixteen patients with non-small cell lung cancer were reviewed with one planning CT and six weekly CBCTs each. A convolutional long short-term memory-based model was applied to predict the weekly deformation of the primary tumor based on the spatial and temporal features extracted from previous weekly CBCTs and target contours. Starting from Week 3, the tumor contour at Week N was predicted by the model based on the input from all the prior weeks (1,2…N-1), and was evaluated against the manually contoured tumor using dice coefficient (DSC) and average surface distance (ASD). Information about the predicted tumor was then used to re-optimize the treatment plan every week. The objectives were set to maximize the dose coverage in the target region while minimizing the toxicity to the surrounding OARs. Dosimetric evaluation of the target and organs at risk was performed on four cases, whereby we compared the conventional plan (ignoring shrinkage) to the shrinkage-based plan.
Results: Average DSC and ASD between the predicted tumor and the actual tumor for Week 3, 4, 5, 6 were (0.81, 0.80, 0.76, 0.78) and (1.72, 1.64, 2.04, 1.97) mm, respectively, which are significantly superior to the scores of (0.70, 0.68, 0.66, and 0.63) and (2.81, 3.22, 3.69, and 3.74) mm between the actual and original volume. While target coverage metrics were maintained for the re-optimized plans, lung mean dose dropped by 2.32, 2.38, 0.36 and 1.25 Gy for four sample cases when compared to the original plan.
Conclusion: We developed a deep-learning-based model for tumor shrinkage prediction. The proposed framework produces reasonable tumor contours and adaption can maintain target coverage while reducing the mean lung dose.
Modeling, Radiation Therapy, Image-guided Therapy