{"id":1006,"date":"2026-02-15T16:57:37","date_gmt":"2026-02-15T16:57:37","guid":{"rendered":"https:\/\/els-journal.net\/wp\/about\/"},"modified":"2026-02-15T16:58:56","modified_gmt":"2026-02-15T16:58:56","slug":"30101-near-instantaneous-cardiovascular-event-prediction-using-multimodal-deep-learning","status":"publish","type":"page","link":"https:\/\/els-journal.net\/wp\/?page_id=1006","title":{"rendered":"30101 Near-instantaneous Cardiovascular Event Prediction Using Multimodal Deep Learning"},"content":{"rendered":"\n\n\n

Vol. 30, No. 1 – June 2026<\/h3>\n

Near-instantaneous Cardiovascular Event Prediction Using Multimodal Deep Learning<\/h3>\n

https:\/\/doi.org\/10.53314\/ELS2630003A<\/h5>\n
Maytham Kareem Naeem Al-Hasooni, Amenah Y. Abdzaid, Noor. H Hadi, and Hassan Falah Fakhruldeen<\/h5>\n
Abstract<\/b><\/h5>\n
This study introduces a new perspective into deep learning in the light of a multimodal approach: cardiovascular <\/span>events can be predicted, using real-time data of physiological signals <\/span>in collaboration with metadata related to the patient. Electronic <\/span>Health Records (EHR) are digital versions of patients\u2019 medical <\/span>histories, while Multilayer Perceptron (MLP) and Convolutional <\/span>Neural Network (CNN) are deep learning architectures designed <\/span>for processing structured data and spatial\/temporal patterns, respectively. <\/span>A hybrid neural network model is designed that allows <\/span>taking, as input from the CNN, the 12-lead ECG signals, while an <\/span>MLP processes patient demographic and clinical features. It is <\/span>designed to simultaneously process temporal ECG patterns and <\/span>static patient characteristics for all-rounded cardiovascular risk <\/span>assessment. In this work, our dataset consisted of 17,441 ECG recordings <\/span>per patient, each being a 12-channel signal sampled on <\/span>500-time points and patient metadata like age, sex, and weight. <\/span>Our architecture has two specialised components: the proposed <\/span>SignalCNN to process the waveforms including two convolutional <\/span>layers with batch normalization and dropout as regularization and <\/span>MetaMLP processing patient metadata. These combined features <\/span>are then fed into a classifier to enable multi-label prediction of five <\/span>common cardiovascular conditions. The model yielded very promising <\/span>results and performed very robustly with an overall validation <\/span>accuracy of 85.19% after 15 epochs of training. The training <\/span>was improving smoothly for both training and validation metrics, <\/span>while the validation loss decreased from 0.4298 to 0.3484, which is <\/span>indicative of good generalization. The model was very stable in its<\/span><\/h5>
training without showing any hint of overfitting thanks to strategic dropout and batch normalization. This work will contribute to <\/span>cardiovascular healthcare with a real-time, automated system that <\/span>can be used for the early detection of cardiac events. The approach <\/span>is multimodal, offering more nuanced predictions by including instantaneous<\/span><\/h5>
physiological signals, together with patient-specific factors. This may enable earlier and more accurate clinical assessment <\/span>of cardiovascular risk.<\/span><\/h5>\n
Full text: <\/a><\/h5>\n\n\n\n\n\"cited<\/a>\n\n\n\n\nGoogle Scholar Citations <\/a>\n\n\n\n\n
<\/span> <\/center>