Multi Modal Deep Learning Based on Feature Attention for Prediction of Blood Clot Elasticity

Abstract

Blood clot is formed inside a blood vessel with various reasons. Carotid artery is major blood vessel in the neck that supply blood to the brain. If blood clot is harden in carotid artery, blood clot of carotid artery can block the blood vessel and make narrowed blood vessel. Therefore, it is essential to predict the coagulation of blood clot in blood vessels. In this paper, we propose the method to determine the coagulation progress of blood clot. We use different two data which are waveform of blood clot and frequency spectra data obtained by applying the Fourier transform to the waveform data. And then feature vectors are extracted from two different data. We apply an encoder block network for waveform data and propose a feature attention network for frequency spectra data. The extracted feature vectors are classified into 3 stages of coagulation progress through multi-modal deep learning. Through the proposed method, we show a meaningful result with an accuracy of 98% in determining the stage of coagulation of blood clot.