Purpose: A deep learning-based fluence map prediction network (FMPN) was developed for predicting fluence maps from desired dose distribution. The network was trained with clinical head and neck (HN) VMAT plans. This work is to investigate the performance of FMPN for various clinical scenarios apart from the training data.
Methods: The FMPN which maps projections of dose to fluence maps was trained only with clinical HN full-arc VMAT plans. To study the generalizability of FMPN, we designed three tests, each of which has a feature different from the training scenario. First, we test the trained network on a different treatment site, prostate. Second, we test the network on new delivery modalities, with three partial-arc VMAT plans and one IMRT plans. Third, we test the network using plans with different degrees of modulation (DOM), to see how well it predicts fluence maps for doses in a wide spectrum of DOM. The generalizability was quantified by comparing the dose of FMPN-predicted fluence maps to the ground truth dose (the desired dose, input of FMPN) using 3D Gamma passing rate with 3mm/3% criteria.
Results: In the first test, FMPN achieved a prediction accuracy in prostate site (99.06±1.75%) as high as in training site (HN, 98.06±2.64%). In the second test, performance was excellent on partial-arc VMAT (99%-89%) and degraded on IMRT (87% with 5mm/5% criteria), which was reasonable because IMRT is too far away from training data (full-arc VMAT) from data distribution point of view. In the third test, FMPN can predict fluence map for plans with various DOM/quality (95%-97%) and accuracy doesn’t vary with DOM.
Conclusion: The FMPN trained on HN full-arc VMAT plans could generalize well to other treatment site, delivery modality and various DOM, showing great potential for clinical use.
Funding Support, Disclosures, and Conflict of Interest: R01 CA235723, R01 CA218402
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation