Click here to


Are you sure ?

Yes, do it No, cancel

Automatic IMRT Planning Via Static Field Fluence Prediction (AIP-SFFP): A Deep Learning Method for Real-Time Prostate Treatment Planning

X Li1*, J Zhang1 , Y Sheng1 , Y Chang1 , F Yin1 , Y Ge2 , Q Wu1 , C Wang1 , (1) Duke University Medical Center, Durham, NC, (2) University of North Carolina at Charlotte, Charlotte, NC


(Monday, 7/15/2019) 1:45 PM - 3:45 PM

Room: Stars at Night Ballroom 2-3

Purpose: To develop a deep learning algorithm of Automatic IMRT Planning via Static Field Fluence Prediction (AIP-SFFP) for real-time prostate treatment planning.

Methods: AIP-SFFP generates a prostate IMRT plan through predictions of fluence maps using the patient anatomy. This is achieved without inverse optimization. AIP-SFFP centralizes a custom-build deep learning network, Dense-Res Hybrid Network (DRHN), which contains both DenseNet and ResNet implementations in a cascade architecture. Predictions from DRHN are imported to Eclipseᵀᴹ system for dose calculation and plan generation.AIP-SFFP was demonstrated for prostate IMRT simultaneously-integrated-boost (SIB) planning (58.8Gy/70Gy to PTV58.8/PTV70 in 28fx). Training data was generated from 105 patients using a 9-beam field template on a knowledge-based planning (KBP) platform based on Eclipseᵀᴹ scripting interface (ESAPI). The following images at each field angle were stacked as inputs for DRHN training: 1) 2D contour projections of PTVs and organs-at-risk (bladder/rectum); 2) digital reconstructed radiographs (DRRs) of CT attenuation coefficients. 10-fold validation was implemented during training. 7 patients were used as independent tests of AIP-SFFP. The generated plans were evaluated by key dosimetric parameters derived from institutional guidelines.

Results: After dose normalization (PTV70 V70Gy=95%), all 7 AIP-SFFP test plans achieved excellent target coverages (PTV58.8 V58.8Gy=98.3±1.8%). Isodose distributions were conformal outside of PTVs with acceptable heterogeneity inside PTV70. 3D max dose values were D0.1cc=106.5±0.5%. Maximum dose to rectum (D0.1cc=72.4±0.6Gy) and bladder (D1cc=71.6±1.1Gy) showed excellent organs-at-risk sparing. V70Gy and V65Gy of rectum and bladder from all 7 plans also met institutional guidelines. Each test plan was generated with 15 seconds or less including prediction and dose calculation. This indicates the feasibility of real-time planning.

Conclusion: AIP-SFFP was successfully developed. Prostate IMRT planning via AIP-SFFP demonstrated good overall plan qualities and real-time efficiency. Holding great promises for clinical application, AIP-SFFP will be investigated for additional clinical sites.


Modeling, Treatment Planning, Radiation Therapy


TH- External beam- photons: Development (new technology and techniques)

Contact Email