Clinical DVH metrics as a loss function for 3D dose prediction in head and neck radiotherapy
Authors: Ruochen Gao, Marius Staring, Frank Dankers
Organizations: Division of Image Processing, Department of Radiology, Leiden University Medical Center, Leiden, the Netherlands. · Department of Radiation Oncology, Leiden University Medical Center, Leiden, the Netherlands.
Purpose: Deep-learning-based three-dimensional (3D) dose prediction is widely used in automated radiotherapy workflows. However, most existing models are trained with voxel-wise regression losses, which are poorly aligned with clinical plan evaluation criteria based on dose-volume histogram (DVH) metrics. This study aims to develop a clinically guided loss formulation that directly optimizes clinically used DVH metrics while remaining computationally efficient for head and neck (H&N) dose prediction. Methods: We propose a clinical DVH metric loss (CDM loss) that incorporates differentiable D-metrics and surrogate V-metrics, together with a lossless bit-mask region-of-interest (ROI) encoding to improve training efficiency. The method was evaluated on 174 H&N patients using a temporal split (137 training, 37 testing). Results: Compared with MAE- and DVH-curve based losses, CDM loss substantially improved target coverage and satisfied all clinical constraints. Using a standard 3D U-Net, the PTV Score was reduced from 1.544 (MAE) to 0.491 (MAE + CDM), while OAR sparing remained comparable. Bit-mask encoding reduced training time by 83% and lowered GPU memory usage. Conclusion: Directly optimizing clinically used DVH metrics enables 3D dose predictions that are better aligned with clinical treatment planning criteria than conventional voxel-wise or DVH-curve-based supervision. The proposed CDM loss, combined with efficient ROI bit-mask encoding, provides a practical and scalable framework for H&N dose prediction.
Figures & tables
Figure 1: Workflow of a typical deep-learning-based H&N dose prediction method. The planning CT and masks of the PTVs and OARs are provided as inputs to a U-shaped neural network to generate a 3D voxel-wise dose distribution. In H&N cases, the PTVs are typically assigned two or three dose levels, while the number of OARs is relatively large compared to other treatment sites.
Parameter
Definition
Dx%
Minimum dose received by x % of the ROI volume.
Dxcc
Minimum dose received by the hottest x cc of the ROI.
Dmax
Maximum dose within the ROI.
Dmin
Minimum dose within the ROI.
Dmean
Mean dose within the ROI.
Vx%
Volume percentage of an ROI receiving at least x % of the prescription dose.
Table 1: Definitions of common D– and V–metrics in H&N RT.
ROI
Metric
Aim
Constraint
PTV54.25
V95%
≥98%
PTV54.25
Dmean
≤102%
PTV70
V95%
≥98%
PTV70
D0.03cc
≤107%
≤110%
PTV70
Dmean
≤102%
SpinalCord
D0.03cc
≤50Gy
Table 2: The H&N RT clinical plan evaluation template used in our institution. The PTVs are denoted by their prescribed dose levels (e.g., PTV54.25 for 54.25 Gy and PTV70 for 70 Gy). ROIs with “L/R” denote paired organs and are evaluated independently for the left and right sides. Each ROI is evaluated using one or more dose metrics. “Aim” represents the desired planning goal, while “Constraint” denotes a mandatory criterion. The template is implemented via a JSON file.
Figure 2: (a) The 3D dose distribution within a ROI is flattened into a one-dimensional dose array {di}i=1N . After a descending sorting operation, we get d1↓≥d2↓≥⋯≥dN↓ . A D–metric , such as Dx% , is obtained by retrieving the dose value dk↓ at the index k that corresponds to the top x% portion of the ROI volume in the sorted dose array. (b) Using the same flattened dose array {di}i=1N , a V–metric such as Vx% is computed by applying a dose threshold equal to x% of the prescription dose, summing the number of voxels that exceed this threshold, and computing the corresponding volume percentage. For example, V95% of PTV70 denotes the fraction of the PTV70 volume receiving at least 95% of the 70 Gy dose level.
Figure 3: Multiple individual ROI masks are losslessly encoded into a single integer bit-mask, where each bit position represents a specific ROI (illustrated in pink). Because the raw bit-mask values span a wide numerical range, direct visualization is not intuitive. Therefore, the bit-mask is visualized in pseudo-color, with colors assigned solely based on their unique values for display purposes. During decoding, the binary mask for ROI i is obtained by testing whether the (i−1) -th bit of the bit-mask is active, implemented as a bitwise AND operation on GPU between the bit-mask and (1≪(i−1)) , where the left-shifted single-bit mask (1≪(i−1)) is illustrated in yellow. A nonzero result indicates voxel membership in ROI i .
Loss function
PTV Score (%) ↓
OAR Score (Gy) ↓
Dose Score (Gy) ↓
MAE
1.544 ± 1.188
2.103 ± 0.704
1.300 ± 0.229
MAE + DVH
0.992 ± 0.466
2.162 ± 0.518
1.385 ± 0.217
MAE + DVH + CDM
0.567 ± 0.300
1.933 ± 0.481
1.389 ± 0.228
MAE + CDM (proposed)
0.491 ± 0.250
1.999 ± 0.497
1.370 ± 0.245
Table 3: Comparison of model performance under different loss-function configurations. Lower scores indicate better performance.
Figure 4: Boxplots of dose evaluation metrics for the clinically most relevant ROIs for different loss function configurations. (a) Boxplots for PTVs. (b) Boxplots for OARs. The corresponding planning aims and dose constraints are also shown (aims are clinically regularly violated due to proximity of the tumor). Parotid and submandibular glands were categorized as ipsilateral (Ipsi) or contralateral (Contra) according to received dose. Statistical significance was tested using a two-tailed Wilcoxon signed-rank test. *: p≤ 0.05, ns = not significant.
Figure 5: Axial visualization of dose distributions under different loss function configurations. The red contour denotes the PTV54.25 , while the blue contour represents the PTV70 . A colormap routinely used in clinical practice is employed to enhance visual interpretability.
α(PTV54.25)
α(PTV70)
PTV Score (%) ↓
OAR Score (Gy) ↓
Dose Score (Gy) ↓
100
100
0.564 ± 0.287
1.919 ± 0.502
1.349 ± 0.228
209
176
0.491 ± 0.250
1.999 ± 0.497
1.370 ± 0.245
300
300
0.500 ± 0.276
1.961 ± 0.491
1.346 ± 0.226
400
400
0.563 ± 0.298
1.988 ± 0.475
1.352 ± 0.214
500
500
0.563 ± 0.308
1.993 ± 0.452
1.327 ± 0.204
Table 4: Ablation study on the parameter α . Lower scores indicate better performance.
Method (3D U-Net baseline)
Peak GPU Memory (GB)
Epoch Duration (s)
Without bit–mask
20.50
241
With bit–mask (proposed)
19.57
43
Table 5: Ablation experiment on the 3D U-Net baseline comparing computational efficiency with and without the bit–mask ROI encoding.
Model
PTV Score (%) ↓
OAR Score (Gy) ↓
Dose Score (Gy) ↓
SwinUNETR Hatamizadeh et al. (2021)
0.699 ± 0.292
2.035 ± 0.566
1.461 ± 0.253
C3D Liu et al. (2021)
0.497 ± 0.314
1.945 ± 0.473
1.289 ± 0.192
MedNeXt Roy et al. (2023)
0.496 ± 0.251
2.154 ± 0.571
1.395 ± 0.204
Pyfer Gheshlaghi et al. (2024)
0.491 ± 0.287
2.019 ± 0.517
1.290 ± 0.207
3D UNet (baseline) Falk et al. (2019)
0.491 ± 0.250
1.999 ± 0.497
1.370 ± 0.245
Table 6: Comparison of dose prediction performance across different network architectures using the MAE + CDM loss configuration. Lower scores indicate better performance.
Background: Manual delineation of target volumes in head and neck cancer (HNC) remains a significant bottleneck in radiotherapy planning, characterized by high inter-observer variability and time consumption. This study evaluates the integration of a Volume-Aware (VA) Dice loss function into a self-configuring deep learning framework to enhance the auto-segmentation of primary tumors (PT) and metastatic lymph nodes (LN) for adaptive MR-guided radiotherapy. We investigate how volume-sensitive weighting affects the detection of small, anatomically complex nodal metastases compared to conventional loss functions. Methods: Utilizing the HNTS-MRG 2024 dataset, we implemented an nnU-Net ResEnc M architecture. We conducted a multi-label segmentation task, comparing a standard Dice loss baseline against two Volume-Aware configurations: a "Dual Mask" setup (VA loss on both PT and LN) and a "Selective LN Mask" setup (VA loss on LN only). Evaluation metrics included volumetric Dice scores, surface-based metrics (SDS, MSD, HD95), and lesion-wise binary detection sensitivity and precision. Results: The Selective LN Mask configuration achieved the highest LN Volumetric Dice Score (0.758 vs. 0.734 baseline) and significantly improved LN Lesion-Wise Detection Sensitivity (84.93% vs. 81.80%). However, a critical trade-off was observed; PT detection precision declined significantly in the selective setup (63.65% vs. 81.27%). The Dual Mask configuration provided the most balanced performance across both targets, maintaining primary tumor precision at 82.04% while improving LN sensitivity to 83.46%. Conclusions: A volume-sensitive loss function mitigated the under-representation of small metastatic lesions in HNC. While selective weighting yielded the best nodal detection, a dual-mask approach is required in multi-label tasks to maintain segmentation accuracy for larger primary tumor volumes.
Sogand Beirami, Zahra Esmaeilzadeh, Ahmed Gomaa +9
Automated treatment planning (ATP) aims to reduce the unwarranted inter-planner variability of manual planning, while improving treatment quality and efficiency. Most ATP pipelines rely on the prediction of the achievable three-dimensional dose distribution for a given anatomy and the subsequent optimization of machine parameters to reproduce it. Diffusion-based methods are largely slice-wise due to the high dimensionality of radiotherapy data, and assume a one-to-one mapping between anatomy and optimal dose despite the multi-objective nature of treatment planning. They are evaluated on their ability to match reference doses and rarely on their integration in ATP frameworks. We present LaDose, a latent diffusion framework for fully volumetric dose prediction. An autoencoder learned on dose distributions, trained so that perturbed latents decode to plausible doses, compresses dose maps 128-fold and enables 3D diffusion on a single GPU. Prescription values are represented as exact scalar vectors that bypass the image encoder, avoiding resampling and latent compression. Multiscale classifier-free guidance allows for the modulation of clinical trade-offs at inference, without retraining or additional labeled data. On 900 clinically delivered head-and-neck volumetric modulated arc therapy plans from five centers, LaDose reached a median absolute error of 1.18 Gy (1.4--1.8% of the prescription) and was the only one of six frameworks free of systematic bias with respect to the reference distribution. Guidance scales acted predictably on target coverage and organ-at-risk sparing. Integrated with direct aperture optimization, the pipeline generated plans in 15.2 minutes; in a blinded review, senior clinicians rated 14 of 15 plans acceptable without editing.
Automating prostate radiotherapy treatment planning is dosimetrically complex, particularly for extreme hypofractionated regimens. In this study, we introduce Dose-PlanNet, a physics-guided 3D deep learning architecture designed to predict dose distributions. This model's performance was evaluated on a cohort of patients treated in a prospective trial where two different dose fractionation regimens were employed. Dose-PlanNet achieved comparable target coverage (D95), though statistical analysis revealed a marginal reduction in target homogeneity (p<0.001) offset. However the model achieved statistically significant improvements in high-dose organ-at-risk sparing (p<0.001). When evaluated against strict Prospective Randomized protocol volumetric constraints, automated plans met prespecified clinical acceptance criteria in 11 out of 14 Moderate Hypofraction Arm plans and 9 out of 12 Stereotactic Body Radiation Therapy Arm plans. This pipeline demonstrates that physics-informed deep learning can accelerate radiotherapy workflows while safely maintaining the stringent dosimetric quality required for high-precision clinical deployment.