Click here to

Session: Therapy: External Beam: Automatic Treatment Planning [Return to Session]

A Deep-Learning-Based Dual-Arc VMAT Plan Generation From Patient Anatomy for Prostate Simultaneous Integrated Boost (SIB) Cases

Q Zhu1*, X Li2, Y Ni3, C Wang4, Q Wu5, Y Ge6, F Yin7, Q Wu8, (1) Duke Kunshan University, Kunshan, 32, CN, (2) Duke University Medical Center, Durham, NC, (3) ,,,(4) Duke University Medical Center, Durham, NC, (5) Duke University Medical Center, Durham, NC, (6) University of North Carolina at Charlotte, Charlotte, NC, (7) Duke University, Chapel Hill, NC, (8) Duke University Medical Center, Durham, NC

Presentations

MO-IePD-TRACK 5-3 (Monday, 7/26/2021) 12:30 PM - 1:00 PM [Eastern Time (GMT-4)]

Purpose: To develop a deep-learning algorithm to automatically generate dual-arc VMAT plans using a patient’s 2D integrated CT attenuation coefficients for prostate SIB cases.

Methods: A customized U-net were designed with 4-resolution-step analysis path and a 4-resolution-step synthesis path. The inputs are 2D integrated CT attenuation coefficients and associated structure contours at the gantry position projection. The output prediction of intensity maps generates two sets of MLC sequences for two VMAT arcs with the same collimator angle. The results were then sent to a commercial TPS for plan finalization. This study involves 140 prostate patients who received simultaneously-integrated-boost (SIB) treatment (PTV(58.8)/PTV(70) in 28fx). A total of 129 one-arc VMAT plans (library plans) were used for network training and validation. For the 11 test patients, the predicted intensity maps are divided into two parts from the isocenter. An empirical elliptical Gaussian distribution was multiplied with the predicted intensity maps. These intensity maps were then divided into upper and lower parts to calculate MLC positions. Key dosimetry metrics in the resulting dual-arc VMAT plans (DL plans) were compared against the one-arc AVP-DSP plans in network training.

Results: After dose normalization (PTV(70) V70Gy=95%), some of 12 DL test case plans met institutional clinic guidelines of dose distribution for OAR and normal tissue sparing. The tested plans showed a high consistency. In an example DL plan, OAR parameters were comparable to one-arc library plan (Bladder V70Gy=5.6cc, V40Gy=15.5%; rectum V70Gy=3.9cc, V40Gy=35.9%). Single-arc: (Bladder (V70Gy=6.8cc, V40Gy=19.4%; rectum (V70Gy=2.8cc, V40Gy = 26.3%). 3D max dose results in DL plans (D1cc=114.6% were lower than the one-arc library plan (D1cc=118.9%). On average, DL plan generation takes 30s/case in contrast to the current clinical practice.

Conclusion: Compared to one-arc library plans, the proposed dual-arc VMAT plan generation algorithm has potentials to achieve higher modulation and lower maximum dose.

ePosters

    Keywords

    Radiation Therapy, Treatment Planning

    Taxonomy

    TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation

    Contact Email

    Share: