Henrik Stelling

Dr. Henrik Stelling

orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 February 2026 in Diagnostics
Verified
Journal Article
Open access
PDF file
Ibrahim GülerIbrahim Güler
Ibrahim Güler
Gerrit GriebGerrit Grieb
Gerrit Grieb
Armin KrausArmin Kraus
Armin Kraus
Martin LautenbachMartin Lautenbach
Martin Lautenbach
Henrik StellingHenrik Stelling
Henrik Stelling
Background/Objectives: Multimodal large language models (MLLMs) offer potential for automated fracture detection, yet their diagnostic stability under repeated inference remains underexplored. This study evaluates the diagnostic accuracy, stability, and intra-model consistency of four MLLMs in detecting hand fractures on plain radiographs. Methods: In total, images of hand radiographs of 65 adult patients with confirmed hand fractures (30 phalangeal, 30 metacarpal, 5 scaphoid) were evaluated by four models: GPT-5 Pro, Gemini 2.5 Pro, Claude Sonnet 4.5, and Mistral Medium 3.1. Each image was independently analyzed five times per model using identical zero-shot prompts (1300 total inferences). Diagnostic accuracy, inter-run reliability (Fleiss’ κ), case-level agreement profiles, subgroup performance, and exploratory demographic inference (age, sex) were assessed. Results: GPT-5 Pro achieved the highest accuracy (64.3%) and consistency (κ = 0.71), followed by Gemini 2.5 Pro (56.9%, κ = 0.57). Mistral Medium 3.1 exhibited high agreement (κ = 0.88) despite low accuracy (38.5%), indicating systematic error (“confident hallucination”). Claude Sonnet 4.5 showed low accuracy (33.8%) and consistency (κ = 0.33), reflecting instability. While phalangeal fractures were reliably detected by top models, scaphoid fractures remained challenging. Demographic analysis revealed poor capabilities, with age estimation errors exceeding 12 years and sex prediction accuracy near random chance. Conclusions: Diagnostic accuracy and consistency are distinct performance dimensions; high intra-model agreement does not imply correctness. While GPT-5 Pro demonstrated the most favorable balance of accuracy and stability, other models exhibited critical failure modes ranging from systematic bias to random instability. At present, MLLMs should be regarded as experimental diagnostic reasoning systems rather than reliable standalone tools for clinical fracture detection.

3 June 2026 in Journal of Plastic, Reconstructive & Aesthetic Surgery
Verified
Journal Article
Open access
Ibrahim GülerIbrahim Güler
Ibrahim Güler
Armin KrausArmin Kraus
Armin Kraus
Gerrit GriebGerrit Grieb
Gerrit Grieb
Henrik StellingHenrik Stelling
Henrik Stelling
Accurate assessment of burn depth and total body surface area (TBSA) is critical for clinical decision-making; however, it remains subjective and prone to interobserver variability. Multimodal large language models (MLLMs) are increasingly encountered in clinical contexts, but whether these systems can reliably assess burn images remains unclear. We evaluated four MLLMs (GPT-5.4 Pro, Grok 4.1, Gemini 3.1 Pro, and Claude Opus 4.6) on 50 clinical burn photographs using a repeated-inference design with five independent runs per model. Burn depth classification was assessed in numeric and text-based formats, alongside ordinal TBSA estimation. Performance varied across the models, with burn depth accuracy ranging from 34.0 ± 6.5% to 76.4 ± 6.8% and TBSA accuracy from 32.8 ± 9.4% to 68.4 ± 3.3%. Inter-run reliability (Fleiss' κ) ranged from slight (κ = 0.171) to almost perfect (κ = 0.916), demonstrating response variability not captured by single-query evaluations. Notably, no model combined high accuracy and high reliability, indicating a dissociation between performance and consistency. All models showed a tendency toward overestimation of burn depth, including assignment of fourth-degree burns despite their absence in the dataset. Error direction analysis revealed model-specific and task-dependent biases, including opposing patterns within the same model. Internal consistency between numeric and text classifications was near-perfect (99.6-100%), indicating format-invariant but systematically biased outputs. These findings demonstrate that MLLM performance is characterized by stochastic response instability invisible to single-query evaluations. Such inconsistency for identical inputs represents a fundamental limitation for workflows requiring consistent outputs across repeated evaluations.

18 February 2026 in AI
Verified
Journal Article
Open access
PDF file
Henrik StellingHenrik Stelling
Henrik Stelling
IBIngo Brink
Ingo Brink
Gerrit GriebGerrit Grieb
Gerrit Grieb
Armin KrausArmin Kraus
Armin Kraus
Ibrahim GülerIbrahim Güler
Ibrahim Güler
Background: Large language models (LLMs) have demonstrated strong performance on general medical examinations. Whether this performance translates to highly specialized, subspecialty-level board examinations remains unclear. This study evaluates the accuracy and inter-run stability of contemporary LLMs using authentic European Board of Nuclear Medicine (EBNM) Fellowship Examination material. Methods: Ten LLMs (five proprietary, five open-source) completed 50 EBNM multiple-choice questions across five independent zero-shot runs, resulting in 2500 total inferences. Accuracy was calculated per model across runs. Inter-run reliability was assessed using pairwise Cohen’s kappa coefficients. Pairwise model differences were analyzed using McNemar’s test with Bonferroni correction (α = 0.0011). Results: Mean accuracy ranged from 53.6% to 100.0%, with all models exceeding an illustrative 50% pass threshold. Inter-run reliability varied substantially (κ = 0.370–1.000; mean κ = 0.716). High accuracy did not consistently correspond to high reproducibility. Gemini 2.5 Pro achieved high accuracy (93.6%) but showed the lowest reliability (κ = 0.370), whereas DeepSeek V3.2 demonstrated perfect accuracy and agreement across all runs. No significant correlation between accuracy and reliability was observed (Spearman ρ = 0.394, p = 0.26). Conclusions: LLMs demonstrate strong but heterogeneous performance on high-stakes medical knowledge assessments. Differences in reproducibility highlight the need for multi-run evaluation when considering LLMs for educational or clinical knowledge-support applications and for continued validation using non-disclosed examination material.

15 April 2026 in Toxins
Verified
Journal Article
Open access
PDF file
Ibrahim GülerIbrahim Güler
Ibrahim Güler
Armin KrausArmin Kraus
Armin Kraus
Gerrit GriebGerrit Grieb
Gerrit Grieb
Henrik StellingHenrik Stelling
Henrik Stelling
Botulinum neurotoxin (BoNT) treatment outcomes are commonly assessed through visual evaluation of facial wrinkle patterns, a process that remains inherently subjective despite structured grading systems. This study evaluated whether contemporary multimodal artificial intelligence (AI) systems can identify facial changes associated with BoNT treatment, using region-specific wrinkle patterns as surrogate markers of underlying muscle activity. A dataset of 46 facial images (23 pre-treatment, 23 post-treatment) was analyzed using four multimodal models, each assessed across five independent runs. Models were tasked with classifying treatment state from single images, detecting wrinkle presence in the forehead, glabella, and periorbital regions, and generating exploratory severity scores and age estimates. Two models achieved 100% accuracy in distinguishing pre- from post-treatment images in this dataset, while region-specific wrinkle detection was variable and frequently did not exceed majority-class baselines. Inter-run reliability varied substantially across models. Exploratory wrinkle severity scores showed directional differences between treatment states, whereas apparent age estimates demonstrated minimal systematic variation. These findings suggest that global facial changes associated with BoNT treatment appear to be detectable in model outputs, but region-specific assessment remains limited, underscoring the need for cautious interpretation and further validation.

16 April 2026 in Life
Verified
Journal Article
Open access
PDF file
Henrik StellingHenrik Stelling
Henrik Stelling
Armin KrausArmin Kraus
Armin Kraus
Gerrit GriebGerrit Grieb
Gerrit Grieb
DBDavid Breidung
David Breidung
Ibrahim GülerIbrahim Güler
Ibrahim Güler
The electrocardiogram (ECG) is a central tool in cardiovascular diagnostics, yet interpretation requires expertise and remains subject to variability. Multimodal large language models (MLLMs) have shown emerging capabilities in medical image analysis, but their performance in ECG interpretation remains insufficiently characterized. This study evaluated the diagnostic accuracy and inter-run reliability of five MLLMs across ECG interpretation tasks. Thirteen standard 12-lead ECGs were presented to five models (ChatGPT-5.3, Gemini 3.1 Pro, Claude Opus 4.6, Grok 4.1, and ERNIE 5.0) across five independent runs per case, yielding 2275 task-level assessments. Six categorical interpretation tasks (rhythm, electrical axis, PR/P-wave morphology, QRS duration, ST/T-wave morphology, and QTc interval) were compared with expert-consensus ground truth, while heart rate estimation was evaluated using mean absolute error (MAE). Overall categorical accuracy ranged from 52.3% to 64.9%. QRS duration classification achieved the highest accuracy (66.2–90.8%), whereas ST/T-wave assessment showed the lowest performance (20.0–41.5%). Heart rate MAE ranged from 14.8 to 46.7 bpm. A dissociation between diagnostic accuracy and inter-run reliability was observed across models. These findings indicate that current MLLMs do not achieve clinically reliable ECG interpretation performance and highlight the importance of assessing diagnostic accuracy and inter-run reliability when evaluating artificial intelligence systems in biomedical diagnostics.

2 May 2026 in Life
Verified
Journal Article
Open access
PDF file
Ibrahim GülerIbrahim Güler
Ibrahim Güler
Armin KrausArmin Kraus
Armin Kraus
Gerrit GriebGerrit Grieb
Gerrit Grieb
TSTevfik Satir
Tevfik Satir
Henrik StellingHenrik Stelling
Henrik Stelling
Multimodal large language models (MLLMs) are increasingly considered for cancer diagnostic support, yet their suitability for histopathological image interpretation remains inadequately characterized. We evaluated six contemporary general-purpose MLLMs (Claude Opus 4.6, Claude Sonnet 4.6, Claude Haiku 4.5, ChatGPT 5.3, Grok 4.2, Gemini 3.1 Pro) on 58 paired hematoxylin and eosin (H&E)-stained breast cancer histopathology images (26 malignant, 32 benign) and corresponding nuclei segmentation masks. Each case was classified five times per model under three conditions, image only (IMAGE), mask only (MASK), and both combined (BOTH), yielding 5220 observations. Mean accuracy dropped from 69.4% (IMAGE) to 49.6% (MASK), below the majority-class baseline of 55.2%. Providing the mask together with the image did not improve classification (68.0%), and for ChatGPT 5.3 produced a net loss of 31 correct predictions. Models maintained elevated mean confidence (67.6) under MASK despite near-random accuracy, and reasoning categories shifted in 67.5% of matched case–run pairs between modalities. Under the conditions tested, current general-purpose MLLMs exhibit strong dependence on visual surface features, fail to effectively integrate spatial structural information, and maintain confidence independent of accuracy. These behavioral limitations are directly relevant to the safe deployment of MLLMs in cancer diagnostic workflows.