AI End-to-End Radiation Treatment Planning Under One Second

Simon Arberet; Riqiang Gao; Martin Kraus; Florin C. Ghesu; Wilko Verbakel; Mamadou Diallo; Anthony Magliari; Venkatesan Karuppusamy

eess.IVMar 6, 2026

AI End-to-End Radiation Treatment Planning Under One Second

Authors: Simon Arberet, Riqiang Gao, Martin Kraus, Florin C. Ghesu, Wilko Verbakel, Mamadou Diallo, Anthony Magliari, Venkatesan Karuppusamy,

Organizations: Digital Technology and Innovation, Siemens Healthineers, Princeton, NJ, USA. · Digital Technology and Innovation, Siemens Healthineers, Erlangen, Germany. · Varian Medical Affairs, Siemens Healthineers, Palo Alto, CA, USA.

Abstract

Artificial intelligence-based radiation therapy (RT) planning has the potential to reduce planning time and inter-planner variability, improving efficiency and consistency in clinical workflows. Most existing automated approaches rely on multiple dose evaluations and corrections, resulting in plan generation times of several minutes. We introduce AIRT (Artificial Intelligence-based Radiotherapy), an end-to-end deep-learning framework that directly infers deliverable treatment plans from CT images and structure contours. AIRT generates single-arc VMAT prostate plans, from imaging and anatomical inputs to leaf sequencing, in under one second on a single Nvidia A100 GPU. The framework includes a differentiable dose feedback, an adversarial fluence map shaping, and a plan generation augmentation to improve plan quality and robustness. The model was trained on more than 10,000 intact prostate cases. Non-inferiority to RapidPlan Eclipse was demonstrated across target coverage and OAR sparing metrics. Target homogeneity (HI = 0.10 $\pm$ 0.01) and OAR sparing were similar to reference plans when evaluated using AcurosXB. These results represent a significant step toward ultra-fast standardized RT planning and a streamlined clinical workflow.

Figures & tables

Figure 1: AIRT end-to-end pipeline for AI VMAT plan generation.

Dose Evaluation Engine $\rightarrow$		DL dose		Eclipse Acuros
Structure	Metric	Eclipse	AI	Eclipse	AI
PTV	HI	0.16 ( 0.03 )*	0.11 ( 0.02 )*	0.10 ( 0.01 )*	0.10 ( 0.01 )*
PTV	D98 (Gy)	38.4 ( 0.8 )*	38.8 ( 0.4 )*	39.3 ( 0.2 )	39.3 ( 0.2 )
Bladder	Dmean (Gy)	6.7 ( 3.3 )*	6.9 ( 3.5 )*	7.3 ( 3.5 )*	7.7 ( 3.8 )*
	D50 (Gy)	2.9 ( 2.5 )	3.0 ( 2.7 )	3.3 ( 2.7 )*	3.6 ( 3.0 )*
	D2 (Gy)	34.8 ( 6.8 )	35.1 ( 6.6 )	35.8 ( 6.0 )*	37.0 ( 5.9 )*

Table 1: Comparison of Eclipse and AIRT planning for intact prostate VMAT cases. All Eclipse plans were optimized using AcurosXB and AIRT plans used the DL dose. For fair comparison, dose for both plan types, was calculated using the DL dose and AcurosXB engines. The Homogeneity Index (HI) is defined as

(D_{2\%}-D_{98\%})/D_{50\%}

following the ICRU convention (lower is better). Data are presented as mean (SD). A * indicates

p<0.05

(Wilcoxon signed-rank test).

Panel of Figure 2 — Figure 2: Dose–volume histograms (DVHs) for six cases corresponding to the 0th (minimum), 20th, 40th, 50th (median), 80th, and 100th (maximum) percentiles of PTV size in the validation dataset using Eclipse AcurosXB dose engine.

	Relative metrics				Absolute metrics
	Fluence-domain		Dose-domain		PTV		Bladder		Rectum
Method	PSNR $\uparrow$	SSIM $\uparrow$	MAE $\downarrow$	PTV MAE $\downarrow$	HI $\downarrow$	CI $\sim$	mean $\downarrow$	D2 $\downarrow$	mean $\downarrow$	D2 $\downarrow$
AIRT (-Aug, -DF, -GAN)	29.41	0.944	0.191	1.711	0.162	0.791	6.94	35.48	4.69	28.58
AIRT (-DF, -GAN)	29.68	0.949	0.174	1.569	0.154	0.801	6.71	35.19	4.77	28.82
AIRT (-GAN)	29.67	0.951	0.166	1.293	0.148	0.801	6.93	35.98	4.98	29.43
AIRT (-Aug)	28.70	0.942	0.174	1.167	0.144	0.802	6.76	34.93	5.19	29.39

Table 2: Ablation results for AIRT variants, evaluated on 62 cases of the validation set. Relative metrics are computed in comparison with target plans. Except for the fluence metrics (PSNR and SSIM), all the other metrics are dose-domain metrics, which are computed after leaf sequencing and using the DL dose engine Kraus et al. (2026) . Absolute DVH metrics are reported for PTV, bladder and rectum. “-Aug” indicates that data augmentation was removed, “-DF” that dose feedback was removed, and “-GAN” that adversarial loss was removed. “full” indicates that all the components of the AIRT method are used. Best values per metric are bold. Units: Bladder and Rectum dose metrics are in Gy; PSNR is in dB; SSIM, HI and CI are unitless.

Figure 3: Averaged DVHs (over the 62 cases of the validation dataset) of the AIRT method for various OAR sparing controls. The “(baseline)” planning, in the legend, means that no input OAR sparing control was used (equivalent to $s_{r}=s_{b}=0$ ). $s_{b}=2\%$ in the legend means that the dose feedback mechanism tried to decrease the dose in the bladder by $2\%$ voxel-wise (likewise for the rectum when $s_{r}=2\%$ ) compared to its input dose distribution.

Panel of Figure 4 — Figure 4: Two-stage training of the AIRT end-to-end VMAT planning pipeline. Top: Stage 1 Full pipeline training without adversarial loss. Bottom: Stage 2: Fluence correction network training with adversarial loss.

Appendix figures & tables23 assets

Supplementary material from the paper’s appendix.

Appendix

Name	Type	Definition / Formula
CTV	Target	$\text{Prostate}\cup\big((\text{Prostate}+3\,\text{mm})\cap\text{Seminal Vesicles}\big)$
PTV	Target	$\text{CTV}+3\,\text{mm}$
PTV $\cap$ Rectum	Derived	$\text{PTV}\cap\text{Rectum}$
PTV Ring	Derived	$(\text{PTV}+20\,\text{mm})-(\text{PTV}+5\,\text{mm})$
PTV–CTV	Derived	$\text{PTV}-\text{CTV}$
50% Ring	Derived	$(\text{PTV}+50\,\text{mm})-(\text{PTV}+15\,\text{mm})$

Appendix

Table S1 : Target and organ-at-risk (OAR) structures used in our Prostate RapidPlan.

Figure S1 : Architecture of the DoseProposer network.

Figure S2 : Beam’s eye view (BEV) transform.

Figure S3 : Architecture of the Bev2Fluence network.

Panel of Figure 11 — Figure S4: Two-stage deep learning dose computation. a, The first stage predicts the voxel-wise spherical harmonics coefficients from the planning CT and fluence maps. b, The second computes the 3D dose map from the CT and spherical harmonics coefficients using a MedNeXT network architecture.

Figure S5 : Architecture of Bev2Fluence correction network.

Figure S6 : Architecture of fluence map discriminator network.

Panel of Figure 14 — Figure S7 : Adversarial fluence shaping results. ( Top ) Mosaic view comparing fluence maps generated by ( a ) AIRT (-GAN, i.e. without adversarial loss), ( b ) AIRT (full, i.e. with adversarial loss), and ( c ) Eclipse clinical plans. ( Bottom ) Zoom-in of representative fluence maps highlighing the improved realism achieved with adversarial training. Both views illustrate the discriminator’s impact in enforcing clinically plausible, approximate two-level structure in generated fluence maps.

DL dose engine
Structure	Metric	Eclipse Plans	AIRT Plans	p-value
PTV	HI	0.16 $\pm$ 0.03	0.11 $\pm$ 0.02	$<$ 0.001
PTV	D98 (Gy)	38.4 $\pm$ 0.8	38.8 $\pm$ 0.4	$<$ 0.001
Bladder	Dmean (Gy)	6.7 $\pm$ 3.3	6.9 $\pm$ 3.5	0.01
	D50 (Gy)	2.9 $\pm$ 2.5	3.0 $\pm$ 2.7	0.29
	D2 (Gy)	34.8 $\pm$ 6.8	35.1 $\pm$ 6.6	0.07

Appendix

Table S2 : Comparison of Eclipse and AIRT planning for intact prostate VMAT cases across 4 dose calculation engines. The homogeneity index (HI) is defined as

(D_{2\%}-D_{98\%})/D_{50\%}

following the ICRU convention (lower is better). All metrics are reported as mean

\pm

standard deviation over the validation dataset. P-values are from the Wilcoxon signed-rank test.

Metric	Mean Diff.	P-value	Non-inferior	Margin	N
PTV HI Mean	0.004	0.00	Yes	0.01	62
PTV HI Median	0.004	0.00	Yes	0.01	62
Bladder Mean Dose	0.392	0.00	Yes	1.50	62
Bladder D50	0.346	0.00	Yes	1.50	62
Bladder D2	1.135	0.03	Yes	1.50	62
Rectum Mean Dose	0.031	0.00	Yes	1.50	62

Appendix

Table S3 : Non-inferiority test results comparing our AI planning to Eclipse planning on DVH metrics evaluated using Eclipse AcurosXB dose engine. A metric is considered non-inferior if the one-sided non-inferiority test is statistically significant (p

<

0.05), and the difference does not exceed the predefined margin. Here, the mean difference is reported as the AI metric minus the Eclipse one.

Panel of Figure 17 — Figure S8 : Dose–volume histograms (DVHs) for six cases corresponding to the 0th (minimum), 20th, 40th, 50th (median), 80th, and 100th (maximum) percentiles of PTV size in the validation dataset using DL dose engine.

Panel of Figure 18 — Figure S9 : Dose–volume histograms (DVHs) for six cases corresponding to the 0th (minimum), 20th, 40th, 50th (median), 80th, and 100th (maximum) percentiles of PTV size in the validation dataset using LTBS dose engine.

Panel of Figure 19 — Figure S10 : Dose–volume histograms (DVHs) for six cases corresponding to the 0th (minimum), 20th, 40th, 50th (median), 80th, and 100th (maximum) percentiles of PTV size in the validation dataset using Eclipse AAA dose engine.

Panel of Figure 20 — Figure S11 : DVHs curves accross all the validation patients showing dose distributions for PTV and OARs using DL dose engine. In bold are the mean curves accross patients.

Panel of Figure 21 — Figure S12 : DVHs curves accross all the validation patients showing dose distributions for PTV and OARs using LTBS dose engine. In bold are the mean curves accross patients.

Panel of Figure 22 — Figure S13 : DVHs curves accross all the validation patients showing dose distributions for PTV and OARs using Eclipse AAA dose engine. In bold are the mean curves accross patients.

Panel of Figure 23 — Figure S14 : DVHs curves accross all the validation patients showing dose distributions for PTV and OARs using Eclipse Acuros dose engine. In bold are the mean curves accross patients.

Panel of Figure 24 — Figure S15 : Dose distributions evaluated with the DL dose engine for the median validation case (Case 4, as shown in the corresponding DVH figure). Left = reference dose (Eclipse/target), Right = AIRT predicted dose. CT slices and contours overlaid.

Panel of Figure 25 — Figure S16 : Dose distributions evaluated with the LTBS dose engine for the median validation case (Case 4, as shown in the corresponding DVH figure). Left = reference dose (Eclipse/target), Right = AIRT predicted dose. CT slices and contours overlaid.

Panel of Figure 26 — Figure S17 : Dose distributions evaluated with the Eclipse AAA dose engine for the median validation case (Case 4, as shown in the corresponding DVH figure). Left = reference dose (Eclipse/target), Right = AIRT predicted dose. CT slices and contours overlaid.

Panel of Figure 27 — Figure S18 : Dose distributions evaluated with the Eclipse AcurosXB dose engine for the median validation case (Case 4, as shown in the corresponding DVH figure). Left = reference dose (Eclipse/target), Right = AIRT predicted dose. CT slices and contours overlaid.

Figure S19 : Mean DVHs for the PTV, bladder, and rectum, averaged over the entire validation set. Suppression factors are reported as a percentage of the predicted baseline dose. The “baseline” curve corresponds to no additional OAR control ( $s_{b}=s_{r}=0$ ). The figure shows that parameters $s_{b}$ and $s_{r}$ independently reduce the OAR dose, with an expected tradeoff on the PTV homogeneity, which worsens as OAR sparing increases.

Module	Notes	Compute Time (ms)
DoseProposer pre-processing	Prepare/crop CT and structure tensors	35
DoseProposer	Initial dose prediction	4
BEV Projection (1st pass)	Beam’s eye view (BEV) projection of the dose	56
Bev2Fluence	Dose BEV-to-Fluence network	18
DL Dose Computation	Predicts dose from fluence maps	654
Dose Error Construction ( Err )	Computes 3D dose error map	1

Appendix

Table S4 : Module-wise compute time breakdown for the proposed pipeline.

Oncology

Oncology studies cancer biology, prevention, diagnosis, treatment, and data-driven ways to improve patient care.

See all 258 papers

Automating RT Planning at Scale: High Quality Data For AI Training
Jan 21, 2025Riqiang Gao, Mamadou Diallo, Han Liu +10
Digital Technology and Innovation, Siemens Healthineers · 2Varian Medical Systems, Siemens Healthineers · 3UCSD +2
Learning to Optimize Radiotherapy Plans via Fluence Maps Diffusion Model Generation and LSTM-based Optimization
May 13, 2026Isabella Poles, Simon Arberet, Riqiang Gao +5Learning to Optimize Oncology
Politecnico di Milano, Milan, Italy · Digital Technology and Innovation, Siemens Healthineers, Princeton, NJ, USA · Digital Technology and Innovation, Siemens Healthineers, Erlangen, Germany
Dose-PlanNet: Physics Based Radiotherapy Dose Prediction with Deep Learning
Aug 27, 2026Ankit Bhattacharjee, Sougata Maity, Santam Chakraborty +1Radiotherapy Dose Prediction Physics-Informed ML
Department of Mechanical Engineering, Indian Institute of Technology Kharagpur · Department of Radiation Oncology, Tata Medical Center Kolkata

Jan 21, 2025J/K move · Enter open · S save

Automating RT Planning at Scale: High Quality Data For AI Training

Riqiang Gao, Mamadou Diallo, Han Liu, Anthony Magliari+9

Digital Technology and Innovation, Siemens Healthineers · 2Varian Medical Systems, Siemens Healthineers · 3UCSD+2

Open paper AskPDF arXiv

Radiotherapy (RT) planning is complex, subjective, and time-intensive. Advances with artificial intelligence (AI) promise to improve its precision and efficiency, but progress is often limited by the scarcity of large, standardized datasets. To address this, we introduce the Automated Iterative RT Planning (AIRTP) system, a scalable solution for generating high-quality treatment plans. This scalable solution is designed to generate substantial volumes of consistently high-quality treatment plans, overcoming a key obstacle in the advancement of AI-driven RT planning. Our AIRTP pipeline adheres to clinical guidelines and automates essential steps, including organ-at-risk (OAR) contouring, helper structure creation, beam setup, optimization, and plan quality improvement, using AI integrated with RT planning software like Varian Eclipse. Furthermore, a novel approach for determining optimization parameters to reproduce 3D dose distributions, i.e. a method to convert dose predictions to deliverable treatment plans constrained by machine limitations is proposed. A comparative analysis of plan quality reveals that our automated pipeline produces treatment plans of quality comparable to those generated manually, which traditionally require several hours of labor per plan. Committed to public research, the first data release of our AIRTP pipeline includes nine cohorts covering head-and-neck and lung cancer sites to support an AAPM 2025 challenge. To our best knowledge, this dataset features more than 10 times number of plans compared to the largest existing well-curated public dataset. Repo: https://github.com/RiqiangGao/GDP-HMM_AAPMChallenge.

Abstract

Oncology

Explore similar work

Automating RT Planning at Scale: High Quality Data For AI Training

Learning to Optimize Radiotherapy Plans via Fluence Maps Diffusion Model Generation and LSTM-based Optimization

Dose-PlanNet: Physics Based Radiotherapy Dose Prediction with Deep Learning