An Artificial Intelligence-Driven Agent for Real-Time Head-and-Neck IMRT Plan Generation using Conditional Generative Adversarial Network (cGAN)

Xinyi Li; Yang Sheng; Jiahan Zhang; Wentao Wang; Fang-Fang Yin; Qiuwen Wu; Yaorong Ge; Q. Jackie Wu; Chunhao Wang

doi:10.1002/mp.14770

paper

An Artificial Intelligence-Driven Agent for Real-Time Head-and-Neck IMRT Plan Generation using Conditional Generative Adversarial Network (cGAN)

arXiv:2009.12898 · doi:10.1002/mp.14770

Abstract

Purpose: To develop an Artificial Intelligence (AI) agent for fully-automated rapid head and neck (H&N) IMRT plan generation without time-consuming inverse planning.$$ $M e t h o d s : T hi s A I a g e n tw a s t r ain e d u s in g a co n d i t i o na l G e n er a t i v e A d v er s a r ia l N e tw or k a r c hi t ec t u r e . T h e g e n er a t or, P y r a N e t, i s an o v e l D ee p L e a r nin g n e tw or k t ha t im pl e m e n t s 28 c l a ss i c R es N e t b l oc k s in p y r ami d - l ik eco n c a t e na t i o n s . T h e d i scr imina t or i s a c u s t o mi z e d 4 - l a y er D e n se N e t . T h e A I a g e n t f i r s t g e n er a t esc u s t o mi z e d 2 D p r o j ec t i o n s a t 9 t e m pl a t e b e aman g l es f r o m 3 D C T v o l u m e an d s t r u c t u r eso f a p a t i e n t . T h ese p r o j ec t i o n s a r e t h e n s t a c k e d a s 4 D in p u t so f P y r a N e t, f r o m w hi c h 9 r a d ia t i o n f l u e n ce ma p s a r e g e n er a t e d s im u l t an eo u s l y . F ina l l y, t h e p r e d i c t e df l u e n ce ma p s a r e im p or t e d in t o a co mm er c ia l t r e a t m e n tpl annin g sy s t e m (T P S) f or pl anin t e g r i t y c h ec k s . T h e A I a g e n tw a s b u i l t an d t es t e d u p o n 231 or o p ha r y n g e a l pl an s f r o ma T P S pl an l ib r a r y . O n l y t h e p r ima r y pl an s in t h ese q u e n t ia l b oos t r e g im e w er es t u d i e d . A c u s t o mi z e d H a r r w a v e l e tl oss w a s a d o pt e df or f l u e n ce ma p co m p a r i so n . I so d ose d i s t r ib u t i o n s in t es t A I pl an s an d T P S pl an s w er e q u a l i t a t i v e l y e v a l u a t e d . K ey d os im e t r i c m e t r i cs w er es t a t i s t i c a l l y co m p a r e d .$ $$ Results: All test AI plans were successfully generated. Isodose gradients outside of PTV in AI plans were comparable with TPS plans. After PTV coverage normalization, $D_{mean}$ of parotids and oral cavity in AI plans and TPS plans were comparable without statistical significance. AI plans achieved comparable $D_{max}$ at 0.01cc of brainstem and cord+5mm without clinically relevant differences, but body $D_{max}$ was higher than the TPS plan results. The AI agent needs ~3s per case to predict fluence maps.$$$$ Conclusions: The developed AI agent can generate H&N IMRT plans with satisfying dosimetry quality. With rapid and fully automated implementation, it holds great potential for clinical applications.

Withdraw for internal review