EUVIMEDEUVIMED entdecken
Crossref

SViM Transformer with Landmark Guided Multi Scale Cross Attention for Automated Facial Paralysis Severity Grading

S. Kiruthiga, C.Anbuananth, V. Dhilipkumar

VFAST Transactions on Software Engineering · 19.06.2026 · 14 (2) · 240-262

DOI: 10.21015/vtse.v14i2.2450 ↗

Kurz gefasst

Fachartikel in der Zeitschrift VFAST Transactions on Software Engineering, online erschienen am 19.06.2026. Verfasst von S. Kiruthiga und 2 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The assessment of the severity of facial paralysis is a daunting task due to the complex nature of muscular movement, variation in the symmetry of the face, and facial expression fluidity. The traditional methods do not sufficiently reflect the global shape of the face as well as the local deformations because they mostly use handcrafted features or single model prediction. To address these shortcomings, we introduce a new SViM (Swin-Vision and MaxViT) fusion architecture which will combine several variants of transformers, thus leveraging the complementary strengths of each model. The model uses 64 point facial landmark detection to obtain facial feature vectors to reflect facial asymmetry and movement, which are important parameters of the amount of paralysis symptoms. Combining global attention process and local feature extraction strategy, SViM can have great severity-grading capabilities on diverse patient data. The experimental findings show that the fusion methodology used in this paper is superior to the conventional CNNs and transformers excel in terms of accuracy and inference latency. Such transformer models have been integrated to enable effective real-time analysis, especially in terms of their implementation in the clinical environment. According to this work, SViM is a reliable AI-assisted tool to objectively measure facial paralysis with an accuracy of 98.16%, which is higher than the reported results in the literature, and thus it can be used by clinicians to make diagnostic and rehabilitative decisions.

Zitieren

APA

Kiruthiga, S., C.Anbuananth, & Dhilipkumar, V. (2026). SViM Transformer with Landmark Guided Multi Scale Cross Attention for Automated Facial Paralysis Severity Grading. VFAST Transactions on Software Engineering, 14(2), 240-262. https://doi.org/10.21015/vtse.v14i2.2450

Vancouver

Kiruthiga S, C.Anbuananth, Dhilipkumar V. SViM Transformer with Landmark Guided Multi Scale Cross Attention for Automated Facial Paralysis Severity Grading. VFAST Transactions on Software Engineering. 2026;14(2):240-262. doi:10.21015/vtse.v14i2.2450

BibTeX
@article{Kiruthiga2026,
  author = {Kiruthiga, S. and C.Anbuananth and Dhilipkumar, V.},
  title = {SViM Transformer with Landmark Guided Multi Scale Cross Attention for Automated Facial Paralysis Severity Grading},
  journal = {VFAST Transactions on Software Engineering},
  year = {2026},
  volume = {14},
  number = {2},
  pages = {240--262},
  doi = {10.21015/vtse.v14i2.2450}
}
Kostenlos starten →