Compact Myocardial Infarction Classification for 3-Lead ECG-Based Temporal Convolution with Encoder-Decoder Network

Myocardial infarction is one of the leading causes of death worldwide, and early detection is crucial for improving patient outcomes. However, while standard 12-lead electrocardiogram are effective for diagnosis, their complexity and extensive electrode setup can limit their practicality in certain...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:The ... International Winter Conference on Brain-Computer Interface s. 1 - 6
Hlavní autoři: Kim, Ray, Choi, Insung, Kim, Gyung Chul, Song, Hee Seok, Lee, Minji, Kim, Yun Kwan
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 24.02.2025
Témata:
3-lead electro-cardiogram signals
convolutional neural network
Convolutional neural networks
Data models
Deep learning
Electrocardiography
myocardial infarction detection
Myocardium
Performance evaluation
Representation learning
Sensitivity
Wearable devices
ISSN:2572-7672
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Myocardial infarction is one of the leading causes of death worldwide, and early detection is crucial for improving patient outcomes. However, while standard 12-lead electrocardiogram are effective for diagnosis, their complexity and extensive electrode setup can limit their practicality in certain clinical settings. To overcome these limitations, this study utilizes 3-lead ECG data collected from leads V1, V3, and V5. Building upon this, we propose a deep learning model that combines temporal convolutional networks with an encoder-decoder convolutional neural network featuring skip connections to classify myocardial infarction. Specifically, each lead is independently processed through dedicated temporal convolutional network blocks to capture temporal patterns, followed by an encoder-decoder convolutional neural network for spatial feature learning. When evaluated on the Physikalisch-Technische Bundesanstalt diagnostic electrocardiogram database, our model demonstrated high performance, achieving an F1-score of 0.955 and a sensitivity of 0.978, yielding results that are comparable to or exceed those of models based on 6 or 12-lead ECG configurations. By reducing the number of leads, this approach simplifies data acquisition and improves convenience for both patients and medical professionals, while also enhancing suitability for portable and wearable devices.
ISSN:2572-7672
DOI:10.1109/BCI65088.2025.10931316