Click here to

Session: Deep Learning for Treatment Planning [Return to Session]

IMRT Plans Guided by Novel Artificial Intelligence Neuro-Fuzzy Algorithm

E Cisternas Jimenez12*, F Yin123, (1) Medical Physics Graduate Program, Duke University, Durham, NC, United States of America , (2) Department of Radiation Oncology, Duke University, Durham, NC, United States of America, (3) Medical Physics Graduate Program, Duke Kunshan University, Kunshan, People's Republic of China

Presentations

TU-I345-IePD-F2-2 (Tuesday, 7/12/2022) 3:45 PM - 4:15 PM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 2

Purpose: IMRT is a complex, multi-criteria problem containing various non-linear relationships. There are two critical challenges to IMRT optimization. First, the tuning for the weighting factors (WF) in the objective function. Second, dealing with multiple intentions that physicians may have represented as different dose prescription levels (DPL). Humans drive the IMRT plan to achieve an ideal dose distribution. This is done manually in a tedious and time-consuming process based on trial and error and user experience. This work aims to develop a novel neuro-fuzzy-guided system to automate these challenges.

Methods: An Artificial Intelligence (AI) technique based on an adaptive network fuzzy inference system (ANFIS) was used in this study. ANFIS combines principles of fuzzy inference systems (FIS) and neural networks (NN). FIS maps a given input (WF and DPL) to an output (Optimal Dose Distribution and DVH) using fuzzy logic. The mapping then provides a basis from which decisions can be made during the IMRT optimization process. The NN component is used to increase the exactness of the FIS model. The ANFIS goal is to make the AI-based plan optimization is not a black box, can be explained how it works, and guide the system to reach the anticipated goal.

Results: It was possible to generate a photon treatment plan fulfilling the clinical goals specified in the TG-119 after 15 iterations. It takes between 8-10 min to achieve the optimal dose. It was a reduction of OARs doses, and an increased PTV dose homogeneity. ANFIS showed excellent agreement between the plan goal and the calculated dose with a variation of 0.22% and 4.62% for D95.

Conclusion: It is feasible to automatically tune the weighting factors for an IMRT plan under the guidance of an optimized FIS without human intervention other than providing the treatment plan parameters and set of constraints.

Funding Support, Disclosures, and Conflict of Interest: E. Cisternas acknowledges support from the Fulbright Ph.D. scholarship program and the Chilean National Agency for Research and Development. The authors declare that they have no conflict of interest.

Keywords

Inverse Planning, Optimization, Treatment Planning

Taxonomy

TH- External Beam- Photons: IMRT/VMAT dose optimization algorithms

Contact Email

Share: