Automatic learning-based selection of beam angles in radiation therapy of lung cancer

The treatment of lung cancer using external beam radiation requires an optimal selection of the radiation beam directions to avoid unnecessarily treatment of normal healthy tissues. We introduce an automated beam selection method, based on learning the relations between beam angles and anatomical features. Using a large dataset of clinical plans, we train a random forest regressor to predict beam angle likelihood. We then use an optimization procedure that incorporates inter-beam dependencies and selects the treatment beams. We present validation results, demonstrating the equivalence of automatically-selected beams and the derived radiation therapy plans to the clinical, manually-planned, ground-truth. The proposed method may be a useful clinical tool for reducing the manual planning workload, while sustaining plan quality.