Purposes:An artificial intelligence based approach was developed to optimize beam orientation with the assistance of expert knowledge based on anatomical structures.Method:For each beam orientation a single beam plan is created with beam shaped to the beam's eye-view projection of target volume.For the voxels belonging to both radiation field and region of interested organ,the inverse distances between those voxels and the radiation source are summed as a value called coverage region of interest(CRI),which represents the whole coverage region of interested organ under specific beam.The table of CRI for PTV and OAR are calculated for all beams.Second,an initial plan with a single beam is created,and the dose difference between planned dose and prescribed dose is calculated.Based on the dose difference,the fuzzy inference system calculates the adjustment of CRI and the new beam is selected based on the updated CRI found in the table of CRI.The new beam then replaces the initial beam and the second beam is selected in the same way in the next loop.The optimization procedure will continue until no available beams subjected to the given constraints are found.Result:Two clinical cases with complex geometric distribution of internal organs were tested.The results showed that the IMRT plans with the optimized beam orientations produced improvements of up to 20%for OARs sparing without compromising tumor dose coverage.Conclusion:An effective approach was demonstrated in automating beam orientation selection in treatment planning of IMRT.With the assistant of fuzzy inference system,the significant improvement of OAR dose sparing was achieved while superior PTV dose was preserved.This approach demonstrates an effective way to combine planning experience and clinical knowledge together in improving the dose quality of IMRT plan.