Image-based Quantification of Postural Deviations on Patients with Cervical Dystonia: A Machine Learning Approach Using Synthetic Training Data

Roland Stenger; Sebastian Löns; Nele Brügge; Feline Hamami; Alexander Münchau; Theresa Paulus; Anne Weissbach; Tatiana Usnich

cs.CVMar 27, 2026

Image-based Quantification of Postural Deviations on Patients with Cervical Dystonia: A Machine Learning Approach Using Synthetic Training Data

Authors: Roland Stenger, Sebastian Löns, Nele Brügge, Feline Hamami, Alexander Münchau, Theresa Paulus, Anne Weissbach, Tatiana Usnich, Max Borsche, Martje G. Pauly, Lara M. Lange, Markus A. Hobert, Rebecca Herzog, Ana Luísa de Almeida Marcelino, Tina Mainka, Friederike Schumann, Lukas L. Goede, Johanna Reimer, Julienne Haas, Jos Becktepe, Alexander Baumann, Robin Wolke, Chi Wang Ip, Thorsten Odorfer, Daniel Zeller, Lisa Harder-Rauschenberger, John-Ih Lee, Philipp Albrecht, Tristan Kölsche, Joachim K. Krauss, Johanna M. Nagel, Joachim Runge, Johanna Doll-Lee, Simone Zittel, Kai Grimm, Pawel Tacik, André Lee, Tobias Bäumer, Sebastian Fudickar

Organizations: Institute of Medical Informatics, University of Lübeck, Lübeck, Germany · Lübeck Centre for Rare Diseases, University Medical Center Schleswig-Holstein, Lübeck, Germany · Institute of Systems Motor Science, University of Lübeck, CBBM, Lübeck, Germany · German Research Center for Artificial Intelligence, AI in Medical Image and Signal Processing, Lübeck, Germany · Department of Neurology, University Medical Center Schleswig-Holstein, Campus Lübeck, Lübeck, Germany · Institute of Neurogenetics, University of Lübeck, Lübeck, Germany · Department of Neurology, University Medical Center Schleswig-Holstein, Campus Kiel, Kiel, Germany · Department of Neurology with Experimental Neurology, Movement Disorders and Neuromodulation Unit, Charité-Universitätsmedizin Berlin, Corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health (BIH), Berlin, Germany · Department of Neurology, Park-Klinik Weißensee, Berlin, Germany · Department of Neurology, University Hospital Würzburg, Würzburg, Germany · Department of Neurology, Medical Faculty and University Hospital Düsseldorf, Heinrich-Heine-University Düsseldorf, Düsseldorf, Germany · Department of Neurology, Maria Hilf Clinics, Mönchengladbach, Germany · Department of Neurosurgery, Hannover Medical School, Hannover, Germany · Department of Neurology, Hannover Medical School, Hannover, Germany · Department of Neurology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany · Department of Parkinson’s disease, Sleep and Movement Disorders, University Hospital Bonn, Bonn, Germany · Institute of Music Physiology and Musicians’ Medicine, University of Music, Drama and Media Hannover, Germany · Department of Neurology, TUM Klinikum Rechts der Isar, Munich, Germany · Section for Clinical Research IT (SKFIT), Institute of Medical Biometry and Statistics (IMBS), University of Lübeck, Lübeck, Germany

Abstract

Cervical dystonia (CD) is the most common form of dystonia, yet current assessment relies on subjective clinical rating scales, such as the Toronto Western Spasmodic Torticollis Rating Scale (TWSTRS), which requires expertise, is subjective and faces low inter-rater reliability some items of the score. To address the lack of established objective tools for monitoring disease severity and treatment response, this study validates an automated image-based head pose and shift estimation system for patients with CD. We developed an assessment tool that combines a pretrained head-pose estimation algorithm for rotational symptoms with a deep learning model trained exclusively on ~16,000 synthetic avatar images to evaluate rare translational symptoms, specifically lateral shift. This synthetic data approach overcomes the scarcity of clinical training examples. The system's performance was validated in a multicenter study by comparing its predicted scores against the consensus ratings of 20 clinical experts using a dataset of 100 real patient images and 100 labeled synthetic avatars. The automated system demonstrated strong agreement with expert clinical ratings for rotational symptoms, achieving high correlations for torticollis (r=0.91), laterocollis (r=0.81), and anteroretrocollis (r=0.78). For lateral shift, the tool achieved a moderate correlation (r=0.55) with clinical ratings and demonstrated higher accuracy than human raters in controlled benchmark tests on avatars. By leveraging synthetic training data to bridge the clinical data gap, this model successfully generalizes to real-world patients, providing a validated, objective tool for CD postural assessment that can enable standardized clinical decision-making and trial evaluation.

Figures & tables

Figure 1: Computational pipeline for TWSTRS item estimation. TWSTRS=Toronto Western Spasmodic Torticollis Rating Scale.

Figure 2: Synthetic avatar dataset and clinical examples of cervical dystonia. (A) Examples of synthetic avatar images used for model training. The dataset includes frontal renderings with varying camera perspectives, lighting, backgrounds, clothing, and morphologies across a range of symptom severities. (B) Front and side views of two individuals with cervical dystonia. Left: Characterized by mild torticollis (0.5), laterocollis (0.5), and a prominent lateral shift (0.7). Right: A complex syndrome featuring torticollis (2.5), laterocollis (0.9), and minimal retrocollis (0.5). Reported values represent the average of 10 independent clinical ratings using static items of the Toronto Western Spasmodic Torticollis Rating Scale (TWSTRS).

TWSTRS item	Avatars ( $\alpha$ [CI])	Real images ( $\alpha$ [CI])	CI Overlap
Torticollis	0.67 [0.58-0.74]	0.64 [0.53-0.72]	Yes
Laterocollis	0.47 [0.37-0.56]	0.48 [0.39-0.56]	Yes
Anterocollis/ Retrocollis	0.81 [0.73-0.86]	0.56 [0.46-0.64]	No (avatars better)
Lateral shift	0.27 [0.19-0.35]	0.31 [0.20-0.40]	Yes

Table 1: Inter-rater reliability (Krippendorff’s

\alpha

) of human ratings with 95% bootstrap confidence intervals for synthetic avatar and real patient images.

Figure 3: Performance and clinical correlation of the TWSTRS algorithm. (A) Comparison of algorithm performance (true positive rate and accuracy) against human raters on the avatar benchmark with known ground-truth head deviations. (B) Correlation between TWSTRS estimations and mean clinical ratings on real-world clinical images across five items: torticollis, laterocollis, anterocollis/retrocollis, and lateral shift. Each dot represents a single patient image. TWSTRS = Toronto Western Spasmodic Torticollis Rating Scale.

Figure 4: Predictions of our proposed algorithm on each frame of a video, recorded by the Move2Screen app as an illustrative example use case of the algorithm. The tasks (green areas) are chosen to make symptoms accessible, like the maximal possible rotation, or the task to let the head move into the most comfortable position.

Figure 5: The annotation platform; A synthesized avatar demonstrating compound head and neck movements along the three principal axes of rotation: yaw (horizontal turning), pitch (vertical nodding), and roll (side-to-side tilting). (B) The Col-Cap evaluation interface, which uses interactive sliders to separately rate upper and lower cervical motion in each rotational plane. (C) Clinical scoring using the TWSTRS motor subscale, with categorical ratings for cervical postures including torticollis, laterocollis, anterocollis, retrocollis, shoulder elevation, and sagittal displacement. In this study, only TWSTRS estimations were used.

Score	Torticollis	Laterocollis	Anterocollis/Retrocollis
0	Absent (0°–4°)	Absent (0°–4°)	Absent (0°–4°)
1	Slight (5°–22°)	Mild (5°–15°)	Mild (5°–15°)
2	Mild (23°–45°)	Moderate (16°–35°)	Moderate (16°–35°)
3	Moderate (46°–67°)	Severe ( $>$ 35°)	Severe ( $>$ 35°)*
4	Severe (68°–90°)	-	-

Table 2: TWSTRS scores from angle ranges

Transformation	Key parameter	Probability
Horizontal Flip	-	$p=0.5$
Random stretch and compress	scale range = [0.7, 1.2]	$p=0.4$
Random cutout	min fraction = 0.1, max fraction = 0.4	$p=0.2$
Random rotation	degree range: [ $-10^{\circ}$ , $+10^{\circ}$ ]	$p=0.3$
Random resized crop	scale range = [0.6, 1.0], size = 240 $\times$ 240	$p=0.6$

Table 3: Data augmentation during training time.

Component	Setting
Backbone	EfficientNet - B0, $1\times 224\times 224$ input
Regression head	Fully-connected (FC) $\to$ lateral shift
Loss function	Correlation aware loss: $0.7\times\text{MSE-loss}+0.3\times\text{Concordance Corr. Coef.}$ on training set; L1-norm on validation set
Optimizer	AdamW (lr= $5\times 10^{-4}$ , weight_decay= $1\times 10^{-4}$ )
Scheduler	ReduceLROnPlateau (monitor=val_corr, factor=0.5, patience=10, min_lr= $1\times 10^{-6}$ )
Batch size	32

Table 4: Model, training, and optimization hyper-parameters.

Metric	Category	Human rating	Algorithm
TPR	Torticollis	92%	84%
	Laterocollis	92%	77%
	Anterocollis/Retrocollis	79%	83%
	Lateral Shift	56%	50%
Accuracy	Torticollis	59%	76%
	Laterocollis	57%	66%

Table 5: Evaluation of human and algorithmic ratings.

Study	Torticollis	Laterocollis	Anterocollis/ Retrocollis	Lateral Shift	Sagittal Shift
Nakamura et al.	0.902	0.369	0.181	$-0.50$	0.557
Ye et al.	0.843	0.667	0.701	—	—
Zhang et al.	0.62	0.60	0.59	—	—
Peach et al.	0.79	0.67	0.66	—	—
Our	0.91	0.81	0.78	0.55	—

Table 6: Reported correlations between algorithmic estimates and human ratings on TWSTRS items.

Task	Time
Preparation	7 sec
Instruction	9 sec
With eyes open, looking at the camera	6 sec
Instruction	12 sec
Eyes closed while the subject moves their head in the most comfortable direction, revealing involuntary dystonic tendencies without visual correction.	10 sec
Instruction	21 sec

Table 7: Video-protocol as implemented in Move2Screen.

Medical Diagnosis

Medical diagnosis uses clinical data, tests, and images to identify diseases, assess risk, and guide care.

See all 261 papers

Apr 2, 2026

Abnormal Head Movements in Neurological Conditions: A Knowledge-Based Dataset with Application to Cervical Dystonia

Abnormal head movements (AHMs) manifest across a broad spectrum of neurological disorders; however, the absence of a multi-condition resource integrating kinematic measurements, clinical severity scores, and patient demographics constitutes a persistent barrier to the development of AI-driven diagnostic tools. To address this gap, this study introduces NeuroPose-AHM, a knowledge-based dataset of neurologically induced AHMs constructed through a multi-LLM extraction framework applied to 1,430 peer-reviewed publications. The dataset contains 2,756 patient-group-level records spanning 57 neurological conditions, derived from 846 AHM-relevant papers. Inter-LLM reliability analysis confirms robust extraction performance, with study-level classification achieving strong agreement (kappa = 0.822). To demonstrate the dataset's analytical utility, a four-task framework is applied to cervical dystonia (CD), the condition most directly defined by pathological head movement. First, Task 1 performs multi-label AHM type classification (F1 = 0.856). Task 2 constructs the Head-Neck Severity Index (HNSI), a unified metric that normalizes heterogeneous clinical rating scales. The clinical relevance of this index is then evaluated in Task 3, where HNSI is validated against real-world CD patient data, with aligned severe-band proportions (6.7%) providing a preliminary plausibility indication for index calibration within the high severity range. Finally, Task 4 performs bridge analysis between movement-type probabilities and HNSI scores, producing significant correlations (p less than 0.001). These results demonstrate the analytical utility of NeuroPose-AHM as a structured, knowledge-based resource for neurological AHM research. The NeuroPose-AHM dataset is publicly available on Zenodo (https://doi.org/10.5281/zenodo.19386862).

Saja Al-Dabet, Sherzod Turaev, Nazar Zaki

College of Information Technology, United Arab Emirates University, Al Ain, United Arab Emirates

Topic pending

Ask PDF

Jun 5, 2026

Impact of Synthetic Lesional MR Images in Automated Focal Cortical Dysplasia Detection in Low-Data Scenarios

Background and Purpose: Automated detection of focal cortical dysplasia (FCD) requires large volumes of voxelwise lesion-delineated MRI data, which are difficult to acquire. This study aims to generate synthetic MRI data exhibiting FCD, assess their realism, and evaluate their impact on automated FCD detection, particularly in reducing the need for manual annotations. Methods: T1-weighted (T1w) and T2-weighted Fluid-Attenuated Inversion Recovery (FLAIR) MRI scans from 131 FCD patients and 90 healthy controls from multiple (3) sites were retrospectively studied. Synthetic MRIs were generated by conditioning a generative network on binary FCD masks. Two neuroradiologists identified real images from a random set of 14 real and 14 synthetic scans. Three nnU-Net models were trained to detect FCD using: (i) real-only (35 FCD / 35 controls), (ii) real (35 FCD / 35 controls) plus synthetic augmentation, and (iii) expanded real data (70 FCD / 70 controls). Results: Experts showed limited ability to distinguish real from synthetic images, with classification accuracy of 60% for T1w and 70% for FLAIR (inter-rater agreement kappa = 0.86). Augmenting automated FCD detection with synthetic data increased sensitivity by 8.14% (p = 0.12) and improved model confidence at true lesion sites (0.83 +/- 0.11 to 0.89 +/- 0.12; p = 0.02). The expanded real-data model further improved sensitivity to 73.8% (p < 0.001) and confidence to 0.90 +/- 0.14 (p = 0.01). Conclusion: Conditional generative networks can generate realistic synthetic FCD-MRIs, reducing labeled data needs by approximately 20% while maintaining equivalent sensitivity. Equivalent amounts of real data, when available, remain more effective than synthetic augmentation.

Prabhjot Kaur, Hakim Ouaalam, Sedat Kandemirli +2

Computational Radiology Laboratory, Boston Children’s Hospital, Harvard Medical School, Boston, MA, USA

Synthetic Data Augmentation Medical Image Anomaly Detection Medical Image Synthesis Conditional Generative Modeling Synthetic Data Evaluation

Ask PDF

Aug 6, 2026

Patient Pose Assessment Using a CT-Based Framework for Synthetic Data Generation

An adequate diagnostic quality of radiographs is essential for reliable diagnoses and treatment planning. The patient's pose during radiography is one of the most important factors determining the diagnostic quality. Since patient positioning is difficult and not standardized, an automated AI-based approach using depth images to automatically assess the patient's pose before the radiograph has been taken would be helpful. Due to regulatory hurdles, however, it is difficult in practice to acquire the required depth images and corresponding radiographs. In this paper, we present a framework that can generate such training data synthetically from Computed Tomography scans. We further show that by pretraining on our generated synthetic dataset consisting of 3077 image pairs of upper ankle joints, the pose assessment of real upper ankle joints can be improved by up to 11 percentage points.

Manuel Laufer, Dominik Mairhöfer, Malte Sieren +7

Institute for Neuro- and Bioinformatics, University of Lübeck, Germany · University Medical Center Schleswig-Holstein, Lübeck, Germany · IMAGE Information Systems Europe GmbH, Rostock, Germany +1

Synthetic Data Pretraining Medical Imaging Synthetic Data Generation Medical Image Synthesis

Ask PDF

Abstract

Medical Diagnosis

Explore similar work

Abnormal Head Movements in Neurological Conditions: A Knowledge-Based Dataset with Application to Cervical Dystonia

Impact of Synthetic Lesional MR Images in Automated Focal Cortical Dysplasia Detection in Low-Data Scenarios

Patient Pose Assessment Using a CT-Based Framework for Synthetic Data Generation