arXiv subjects

Johan Sundström

Publications and source records attributed to Johan Sundström.

7 recordsLinked to original sources

arXiv2026

On the role of the tokenizer in ECG transformer models

Tokenization determines both the physiological content presented to an ECG Transformer and the sequence over which attention operates. We compare eight tokenization strategies across Transformer, Informer, Reformer, and FEDformer on the nine-label CPSC2018 classification task. The input projection and principal backbone capacity are controlled to isolate the effect of token construction. Median-beat and HeartLang tokenization achieve mean macro-AUCs of 0.893 and 0.889 across the four backbones, compared with 0.822 and 0.824 for point-wise and patch-wise tokenization. Pooling the two physiology-aware representations yields an 8.2% relative improvement in macro-AUC. They also reduce mean sequence length from 1,250 to 158 tokens and mean peak training memory from 5.21 to 0.27 GB. The results show that aligning tokens with ECG morphology can improve both predictive performance and memory efficiency without increasing backbone capacity. The source code is available on https://github.com/LeeJarvis996/ecg_tokenizer.

cs.LG↗

arXiv2026

How Do Electrocardiogram Models Scale?

While scaling laws have established a fundamental framework for foundation models in natural language processing, their applicability to electrocardiogram (ECG) models remains poorly characterized. Indeed, recent studies do not always yield consistent downstream gains as one increases the model size or pre-training dataset size of ECG models, leaving the exact roles of architectural inductive biases, pre-training paradigms, and expected improvements with size largely unanswered. In this work, we systematically investigate neural and loss-to-loss scaling laws within the ECG domain. By pre-training over $120$ models (ranging from $20$K to $200$M parameters) on the large-scale CODE dataset ($2.3$M records), we decouple the effects of model architecture (ResNet vs. Transformer) and pre-training paradigm, namely supervised learning (SL) versus self-supervised learning (SSL). We found that (i) SL models are data-bottlenecked in-distribution, whereas SSL models scale robustly across both model and data sizes; (ii) for out-of-distribution (OOD) generalization, ResNets are $1.3$ to $2.5$ times more parameter-efficient than Transformers, while SSL is up to $16$ times more data-efficient and achieves up to $7.6$ times higher transfer efficiency than SL on unseen clinical tasks; (iii) across the observed scales, ResNet-based models generally achieve the lowest OOD loss, with SSL dominating on unseen clinical tasks and self-supervised Transformers overtaking at very large model sizes. Our results suggest that the path to effective ECG foundation models lies in the strategic alignment of architecture and paradigm rather than brute-force scaling.

cs.LG↗

arXiv2026

Sparse probabilistic evaluation for treatment planning: a feasibility study in IMPT head & neck patients

Probabilistic evaluation improves the trade-off between target coverage and OAR sparing in IMPT but remains computationally demanding. This study proposes sparse probabilistic evaluation (SPE), a computationally efficient approach integrated into a clinical TPS. Clinical plans of 20 IMPT HNC patients treated in 2024 were included. SPE used a predefined setup and range error grid with Monte Carlo computed dose distributions. Two grid settings were evaluated: the maximum error Emax (3$σ$ or 4$σ$) and the number of setup error points nsetup (7, 33, 123). Accuracy and duration of SPE with each grid were evaluated in the calibration group (5 patients). 1000 treatments with normally distributed random ($σ$ = 1 mm) and systematic ($σ$ = 0.92 mm) setup and range ($σ$ = 1.5%) errors were simulated. The dose distribution of the nearest error point in the grid was assigned to each fraction. Probability distributions derived from SPE were compared with those from a reference based on 35,000 Monte Carlo calculations. The found optimal grid (Emax = 3$σ$, nsetup = 33) was applied to the validation group (15 patients). Accuracy of SPE in the calibration group increased significantly as the number of error points increased from 7 (tavg = 2 minutes) to 33 (tavg = 9 minutes), with no further improvement between 33 and 123 (tavg = 27 minutes) error points. Increasing Emax only improved accuracy for values above the 98th percentile. Applying SPE to the validation group resulted in median errors of 0.02 Gy RBE (range:-0.11 to 0.07) for the 10th percentile of the D99.8%, CTV distribution and 0.0 Gy RBE (range:-0.14 to 0.23) for the 95th percentile of the D0.03cc,SpinalCord Core distribution. Sparse probabilistic evaluation achieves sufficient accuracy while requiring clinically acceptable computation times, paving the way for probabilistic evaluation in clinical practice.

physics.med-ph↗

arXiv2026

Percentile-based probabilistic optimization for systematic and random uncertainties in radiation therapy

Geometric uncertainty can degrade treatment quality in radiation therapy. While margins and robust optimization mitigate these effects, they provide only implicit control over clinical goal fulfillment probability. We therefore develop a probabilistic planning framework using a percentile-based optimization function that targets a specified probability of clinical goal fulfillment. Systematic and random uncertainties were explicitly modeled over full treatment courses. A scenario dose approximation method based on interpolation between a fixed set of doses was used, enabling efficient simulation of treatment courses during optimization. The framework was evaluated on a prostate case treated with volumetric-modulated arc therapy (VMAT) and a brain case treated with pencil beam scanning (PBS) proton therapy. Plans were compared to conventional margin-based and worst-case robust optimization using probabilistic evaluation. For the prostate case, probabilistic optimization improved organ at risk (OAR) sparing while maintaining target coverage compared to margin-based planning, increasing average OAR goal fulfillment probability by 13.3 percentage points and reducing 90th percentile OAR doses by an average of 3.5~Gy. For the brain case, probabilistic optimization improved target minimum dose passing probabilities (e.g., 88\% vs.~22\% for $D_{95}$) and brainstem maximum dose passing probability (70\% vs.~30\%), while maintaining comparable or improved OAR sparing compared to worst-case optimization. Probabilistic optimization enables explicit and interpretable control over goal fulfillment probabilities. Combining full treatment course modeling with efficient approximate dose calculation, the proposed framework improved the trade-off between target coverage and OAR sparing compared to conventional planning approaches in both photon and proton therapy.

physics.med-ph↗

arXiv2025

Shoot-through layers in upright proton arcs unlock advantages in plan quality and range verification

Background: Upright proton therapy with compact delivery systems has the potential to reduce costs for treatments but could also lead to broadening of the beam penumbra due to energy selection close to the patient. Purpose: This study aims at combining upright static proton arcs with additional layers of shoot-through (ST) protons to sharpen the beam penumbra and improve plan quality for such systems. Methods: We examined various treatment plans for a virtual phantom: 3-beam IMPT, static arc (Arc) with/without ST (Arc+ST), and with/without collimation (+Coll). In the virtual phantom three different targets were utilized to study the effect on conformity index (CI), homogeneity index (HI), robustness and mean dose to the phantom volume. The phantom study was complemented with a head-and-neck (H&N) patient case with a similar set of plans. The delivery time for all plans was estimated. A range verification concept that determines residual ranges of the ST protons was studied in simulated scenarios for the H&N case. Results: In the phantom study, the Arc+ST plans show superior CI, HI and target robustness compared to the Arc+Coll plans. For the Arc plans without ST, the collimated plans perform better than the uncollimated plans. On the other hand, for Arc+ST, collimation has little impact on CI, HI and robustness. For the H&N case, similar improvements for Arc+ST can be seen. These results imply that no aperture is needed when combining arcs with ST, which in turn substantially reduces treatment times. The range verification simulation shows that the method is sensitive to detect SPR errors, setup errors and changes in the patient anatomy. Conclusions: Combining proton arcs and ST layers can enhance compact upright proton solutions by improving plan quality. It is also tailored for the inclusion of a fast and straightforward residual range verification method.

physics.med-ph↗

arXiv2025

Partitioning of multiple brain metastases improves dose gradients in single-isocenter radiosurgery

Background: A growing number of cancer patients with brain metastases can benefit from stereotactic radiosurgery (SRS) thanks to recent advances in systemic therapies. With an increasing patient load, single-isocenter treatments on widely available C-arm linear accelerators are an attractive option. However, the planning of such treatments is challenging for multi-target cases due to the island blocking problem, which occurs when the multi-leaf collimator cannot conform to all targets simultaneously. Purpose: We propose a multi-target partitioning algorithm that mitigates excessive exposure of normal tissue caused by the island blocking problem. Methods: The algorithm divides (partitions) the set of targets into subsets to treat with separate arc passes, optimizing both subsets and collimator angles to minimize island blocking. The algorithm was incorporated into a fully automated treatment planning script and evaluated on 20 simulated patient cases, each with 10 brain metastases and 21 Gy prescriptions. It was also retrospectively evaluated on six clinical cases. Results: Partitioning significantly improved the gradient index, global efficiency index, and brain V12Gy compared to simultaneous treatment of all metastases. For example, the average gradient index improved from 5.9 to 3.3, global efficiency index from 0.32 to 0.46, and normal brain V12Gy from 49 cm3 to 26 cm3 between 3 and 9 arcs. The proposed algorithm outperformed baselines in utilizing a limited number of arcs. All target partitioning strategies increased the total number of monitor units (MUs). Conclusions: The dose gradient in single-isocenter VMAT plans can be substantially improved by treating a smaller subset of metastases at a time. This requires more MUs and arcs, implying a trade-off between delivery time and plan quality which can be explored using the algorithm proposed in this paper.

physics.med-ph↗

arXiv2022

ECG-Based Electrolyte Prediction: Evaluating Regression and Probabilistic Methods

Objective: Imbalances of the electrolyte concentration levels in the body can lead to catastrophic consequences, but accurate and accessible measurements could improve patient outcomes. While blood tests provide accurate measurements, they are invasive and the laboratory analysis can be slow or inaccessible. In contrast, an electrocardiogram (ECG) is a widely adopted tool which is quick and simple to acquire. However, the problem of estimating continuous electrolyte concentrations directly from ECGs is not well-studied. We therefore investigate if regression methods can be used for accurate ECG-based prediction of electrolyte concentrations. Methods: We explore the use of deep neural networks (DNNs) for this task. We analyze the regression performance across four electrolytes, utilizing a novel dataset containing over 290000 ECGs. For improved understanding, we also study the full spectrum from continuous predictions to binary classification of extreme concentration levels. To enhance clinical usefulness, we finally extend to a probabilistic regression approach and evaluate different uncertainty estimates. Results: We find that the performance varies significantly between different electrolytes, which is clinically justified in the interplay of electrolytes and their manifestation in the ECG. We also compare the regression accuracy with that of traditional machine learning models, demonstrating superior performance of DNNs. Conclusion: Discretization can lead to good classification performance, but does not help solve the original problem of predicting continuous concentration levels. While probabilistic regression demonstrates potential practical usefulness, the uncertainty estimates are not particularly well-calibrated. Significance: Our study is a first step towards accurate and reliable ECG-based prediction of electrolyte concentration levels.

eess.SP↗