deepPatella. A tool for automatic tracking of Patella Tendon Elongation and stiffness calculation
Loading...
Author (Corporation)
Publication date
2026
Type of student thesis
Course of study
Type
01A - Journal article
Editors
Editor (Corporation)
Supervisor
Parent work
Current Issues in Sport Science
Special issue
DOI of the original publication
Link
Related research data
Series
Series number
Volume
11
Issue / Number
2
Pages / Duration
022
Patent number
Publisher / Publishing institution
Bern Open Publishing
Place of publication / Event location
Edition
Version
Programming language
Assignee
Practice partner / Client
Abstract
Patellar tendon (PT) stiffness is highly load-sensitive and can decrease by up to 29% after 14 days of unloading, impacting force transfer and thereby rehabilitation processes and athletic performance. While automated ultrasound (US) analysis has been applied to the Achilles tendon, PT elongation—an indirect marker of stiffness—still relies on manual and time-consuming frame-by-frame annotation. We present deepPatella, a novel, automated, transformer-based, open-source software with Kalman (KF) filtering to efficiently (and objectively) track PT elongation from US frame sequences. PT elongation was assessed using US (ArtUS, Telemed) with a 6 cm linear probe (LF9-5N60-A3 @7 MHz) fixed over the PT apex and tibial tuberosity during ramped isometric knee extensions. US images were recorded at 50 fps. Elongation was defined as relative displacement of proximal and distal insertions, manually labelled by two raters. The dataset included 17.475 frames from 30 participants (13 males, age=46.5 years (35-56), weight=73.7 kg (53.1-95.1), height=171.3 cm (158.0-188.0)), split into 12.221/5.254 frames (or 27/14 videos) for training/testing. A TransUNet model with sigmoid-based heatmaps, BCE-Dice loss, and ADAM optimizer predicted insertion points; predictions were subsequently smoothed using a Kalman filter. Accuracy was evaluated via Euclidean distance to manual labels and by comparing predicted and Kalman filtered PT elongation against the manually labelled one. In the 14 test videos, PT elongation was 4.8±1.5 mm based on manual annotation, 4.6±2.8 mm from the model prediction, and 4.6 ±2.9 mm with KF. Mean Euclidean errors compared to manual labels were 0.9±0.5 mm for the distal and 1.0±0.4 mm for the proximal insertions, for both model and Kalman. Accordingly, total absolute elongation Euclidean errors were 1.0±0.9 mm for the model and 1.0±1.8 mm for Kalman. Overall, the model predicted PT insertions with high accuracy and KF smoothed but did not improve the predictions. Additionally, using deepPatella, the processing time of per test video was reduced by 90% from 30 minutes (manual labelling) to 3 minutes. Thus, deepPatella enables time-efficient and objective PT elongation analysis in US videos and provides the basis for automated stiffness estimation. The openly available deepPatella software further enables users to compute tendon elongation and stiffness on their own data.
Keywords
Subject (DDC)
Event
Exhibition start date
Exhibition end date
Conference start date
Conference end date
Date of the last check
ISBN
ISSN
2414-6641
Language
Other
Created during FHNW affiliation
Yes
Strategic action fields FHNW
Publication status
Published
Review
peer-reviewed
Open access category
Closed
License
Citation
Romeny Kunstmann, G., Ledergerber, R., Keller, M., Behr, D., Benini, L., Faude, O., Leitner, C., & Ritsche, P. (2026). deepPatella. A tool for automatic tracking of Patella Tendon Elongation and stiffness calculation. Current Issues in Sport Science, 11(2), 22. https://doi.org/10.36950/2026.2ciss022