arXiv subjects

Lukas Zimmermann

Publications and source records attributed to Lukas Zimmermann.

7 recordsLinked to original sources

arXiv2026

PyDoseRT Proton: A GPU Pencil-Beam Engine with a Convolutional Residual-Correction Network for Fast Proton Dose Calculation

Architecture category. Hybrid method: a physics-based analytical pencil-beam (PB) dose engine followed by a 3-D convolutional residual-correction network (RepVGG-U-Net). We addressed the DoseRAD2026 proton dose-prediction task with PyDoseRT Proton, a GPU-accelerated engine implemented in PyTorch and augmented by a learned residual toward Monte Carlo (MC) accuracy. A double-Gaussian PB kernel was calibrated to GATE/Geant4 integrated depth doses in water in two stages: a classical per-energy curve fit, then a gradient-based fit of the full 3-D dose through the PyTorch physics engine as it retains a differentiable execution path for gradient-based optimization of dose-dependent objectives. The engine computes each beamlet on a beam's-eye-view (BEV) lattice with variance-preserving Gaussian splitting, an analytic nuclear halo, and a Fermi-Eyges heterogeneity term, then rotates the result into the patient frame. Additionally, a compact residual U-Net predicts an additive correction in BEV space. It is conditioned on voxelwise material-label embeddings, a discrete energy embedding and spot size. The same model was used for all anatomical sites (thoracic and abdominal). It was trained with a patient-space L1 objective emphasizing the scored high-dose region and multi-scale BEV deep supervision. The submitted CT configuration obtained preliminary-test beamlet MAE 0.0066, image-z IDD distance 0.0025, plan MAE 0.0049, 98.30\% gamma pass rate (1\%/1 mm), and DVH error 0.460.

physics.med-ph↗

arXiv2026

PyDoseRT Photon: Physics-Guided Pencil-Beam Dose Calculation with Neural Priors and Residual Correction for CT and MRI

We present a hybrid, physics-based analytical pencil-beam (PB) dose engine, augmented by two small frozen neural physics priors, followed by a 3-D convolutional residual-correction network (U-Net). We address the DoseRAD2026 (https://doserad2026.grand-challenge.org/) photon dose-prediction task with PyDoseRT Photon, a GPU PB engine implemented in PyTorch and corrected toward Monte Carlo (MC) accuracy in three learned stages of decreasing physical specificity. The engine reproduces the challenge's head-less MC source exactly where it can and models the patient with a beam-quality-indexed pencil kernel evaluated at the field's fluence-weighted radiological depth, and TERMA source scaling at the interaction site. Two tiny neural priors are trained through the frozen engine and then frozen themselves: a 39k-parameter 2-D fluence correction and a 48-parameter lateral heterogeneity correction mixing mass-conserving Gaussian redistribution operators. A compact 3-D U-Net (1.36M parameters) with a sequential refinement branch then predicts, per control point (CP) in the beam's-eye-view (BEV) frame, a bounded multiplicative gain and additive residual from seven channels. All learned stages are zero-initialized, so training starts from the analytical solution. For MRI, an nnU-Net regression model synthesizes a CT that enters the identical pipeline, with consecutively, the same trained corrector as the CT track. Design choices were driven by the challenge ranking, in which runtime carries double weight. The submitted method evaluated on our local CT and MR validation dataset achieved CP MAE 0.0086 and 0.0098, IDD distance 0.0011 and 0.0013, plan MAE 0.0024 and 0.0047, gamma pass rate (1%/1mm) 99.12% and 96.95%, with runtimes of 46s and 49s, respectively.

physics.med-ph↗

arXiv2026

COBRA2026: a large-scale multicenter pelvic cone-beam computed tomography projection dataset

The COBRA2026 dataset is a large-scale, multicenter resource of raw radiotherapy cone-beam computed tomography (CBCT) acquisitions created for the development and evaluation of conventional and learning-based reconstruction and image-correction methods. It contains data from 867 patients undergoing pelvic radiotherapy at six European centers, acquired using Elekta and Varian imaging systems. For each case, the dataset includes raw projection data, acquisition geometry, calibration and correction information, clinically reconstructed CBCT images, and corresponding planning CT images. Vendor-specific files were anonymized and converted into open formats. Planning CT images were deformably registered to the daily CBCT anatomy, and matched projections were simulated using the corresponding acquisition geometry. All cases underwent visual quality control, and cases with substantial processing or registration errors were excluded. The approximately 950 GB dataset is divided into training, validation, and test sets containing 692, 52, and 123 cases, respectively. Projection stacks and volumetric images are provided as compressed MetaImage files, with geometry and metadata supplied in XML and YAML formats. COBRA2026 supports research on full- and sparse-view reconstruction, low-dose imaging, artifact and scatter correction, motion compensation, and synthetic CT generation. The dataset is released under the CC BY-NC 4.0 license, indexed on Zenodo (doi:10.5281/zenodo.21322350), and accompanied by openly available preprocessing and baseline reconstruction code. It also forms the basis of the COBRA2026 reconstruction challenge.

physics.med-ph↗

arXiv2026

Eliminating Registration Bias in Synthetic CT Generation: A Physics-Based Simulation Framework

Supervised synthetic CT generation from CBCT requires registered training pairs, yet perfect registration between separately acquired scans remains unattainable. This registration bias propagates into trained models and corrupts standard evaluation metrics. This may suggest that superior benchmark performance indicates better reproduction of registration artifacts rather than anatomical fidelity. We propose physics-based CBCT simulation to provide geometrically aligned training pairs by construction, combined with evaluation using geometric alignment metrics against input CBCT rather than biased ground truth. On two independent pelvic datasets, models trained on synthetic data achieved superior geometric alignment (Normalized Mutual Information: 0.31 vs 0.22) despite lower conventional intensity scores. Intensity metrics showed inverted correlations with clinical assessment for deformably registered data, while Normalized Mutual Information consistently predicted observer preference across registration methodologies (rho = 0.31, p < 0.001). Clinical observers preferred synthetic-trained outputs in 87% of cases, demonstrating that geometric fidelity, not intensity agreement with biased ground truth, aligns with clinical requirements.

cs.CV↗

arXiv2025

Accelerating Transfer Function Update for Distance Map based Volume Rendering

Direct volume rendering using ray-casting is widely used in practice. By using GPUs and applying acceleration techniques as empty space skipping, high frame rates are possible on modern hardware. This enables performance-critical use-cases such as virtual reality volume rendering. The currently fastest known technique uses volumetric distance maps to skip empty sections of the volume during ray-casting but requires the distance map to be updated per transfer function change. In this paper, we demonstrate a technique for subdividing the volume intensity range into partitions and deriving what we call partitioned distance maps. These can be used to accelerate the distance map computation for a newly changed transfer function by a factor up to 30. This allows the currently fastest known empty space skipping approach to be used while maintaining high frame rates even when the transfer function is changed frequently.

cs.GR↗

arXiv2022

Bringing the Algorithms to the Data -- Secure Distributed Medical Analytics using the Personal Health Train (PHT-meDIC)

The need for data privacy and security -- enforced through increasingly strict data protection regulations -- renders the use of healthcare data for machine learning difficult. In particular, the transfer of data between different hospitals is often not permissible and thus cross-site pooling of data not an option. The Personal Health Train (PHT) paradigm proposed within the GO-FAIR initiative implements an 'algorithm to the data' paradigm that ensures that distributed data can be accessed for analysis without transferring any sensitive data. We present PHT-meDIC, a productively deployed open-source implementation of the PHT concept. Containerization allows us to easily deploy even complex data analysis pipelines (e.g, genomics, image analysis) across multiple sites in a secure and scalable manner. We discuss the underlying technological concepts, security models, and governance processes. The implementation has been successfully applied to distributed analyses of large-scale data, including applications of deep neural networks to medical image data.

cs.LG↗

arXiv2022

OpenKBP-Opt: An international and reproducible evaluation of 76 knowledge-based planning pipelines

We establish an open framework for developing plan optimization models for knowledge-based planning (KBP) in radiotherapy. Our framework includes reference plans for 100 patients with head-and-neck cancer and high-quality dose predictions from 19 KBP models that were developed by different research groups during the OpenKBP Grand Challenge. The dose predictions were input to four optimization models to form 76 unique KBP pipelines that generated 7600 plans. The predictions and plans were compared to the reference plans via: dose score, which is the average mean absolute voxel-by-voxel difference in dose a model achieved; the deviation in dose-volume histogram (DVH) criterion; and the frequency of clinical planning criteria satisfaction. We also performed a theoretical investigation to justify our dose mimicking models. The range in rank order correlation of the dose score between predictions and their KBP pipelines was 0.50 to 0.62, which indicates that the quality of the predictions is generally positively correlated with the quality of the plans. Additionally, compared to the input predictions, the KBP-generated plans performed significantly better (P<0.05; one-sided Wilcoxon test) on 18 of 23 DVH criteria. Similarly, each optimization model generated plans that satisfied a higher percentage of criteria than the reference plans. Lastly, our theoretical investigation demonstrated that the dose mimicking models generated plans that are also optimal for a conventional planning model. This was the largest international effort to date for evaluating the combination of KBP prediction and optimization models. In the interest of reproducibility, our data and code is freely available at https://github.com/ababier/open-kbp-opt.

physics.med-ph↗