Work

Модальная, временная и событийная хрупкость мультимодальной модели искусственного интеллекта при анализе тревог желудочковой тахикардии

Year: 2026
Type: article
Abstract: The paper considers fragility assessment of a multimodal physiological time-series model for classifying true and false ventricular tachycardia alarms. The model is evaluated under several stress profiles: full input, ECG removal, hard and conflict events, temporal window shifts, and calibration checks. The results show that high performance on the full test set may hide modality and temporal fragility, while fine-tuning strategies provide only bounded improvements. An external check on the PhysioNet/CinC Challenge 2015 dataset demonstrates that the evaluation procedure can be applied to an independent ICU alarm source.
Language: Russian
Field: Medicine
Open Access status: diamond
OpenAlex ID: W7225443502
FWCI:
Cites: 16
Cited by:
Related to: 10