An Iraqi student developed an artificial intelligence algorithm to diagnose myocardial infarction, detect heart attacks, and blockages of blood vessels feeding the heart.
The algorithm, developed by Mekatronic engineering student Hussein Qadduri, aims to find changes in recorded ECG signals using long-wearable holter devices that typically last from 24 to 48 hours.
Myocardial infarction is a condition that occurs when a blood vessel in the heart is blocked. This prevents the delivery of enough oxygen-laden blood to the heart muscle and leads to damage over time. It is a common disease in the world and is usually diagnosed using normal ECG and blood analysis to measure protein levels Troponin.
The operation of ECG signals from these devices, noise from the movement of the electrodes, may cause a wrong diagnosis when consulted by a specialist. These signals are usually filtered using traditional noise-canceling techniques at specific frequencies.
In contrast, the new algorithm extracts the ECG signal from random noise without relying on specific frequencies, using the deep learning of the machines. It trains the algorithm on a wide range of data; it contains 160,000 ECGs, each 10 seconds long, European database for academic research.
The algorithm achieved good results in filtering signals from random interference and restoring the real signal, after examining data containing 59,000 ECGs; each ECG was 10 seconds long.
"The new algorithm I have developed identifies changes to myocardial infarction in ECG signals, which are changes in planning signals ranging from ST segment," said Hussein Kadouri of the Observatory of the Future.
"The disease is traditionally diagnosed by the height or low of this part of the ECG, and the second algorithm developed by the patient has achieved good results in diagnosing patient planning with 95.7% accuracy and 97.6% accuracy for no-noise signals. The algorithms were then combined to form a single algorithm capable of diagnosing the disease in signals that contained unexpected random noise and were not trained by the neural network. The algorithm achieved good results in diagnosing the disease accurately 83.2% and accuracy of 79.4% in the diagnosis of sound signals; these rates are higher in cases of natural noise ».
This algorithm can also be used in wearable devices to diagnose myocardial infarction without being reviewed by a specialist doctor, which contributes to the speed and accuracy of diagnosis without human effort in reviewing long-term recorded signals.
"The techniques applied in the research are the deep learning of machines, one of the branches of artificial intelligence, which relies on the training of algorithms using data sets. It requires a complex structure that simulates the neural networks of the human brain, to understand patterns even with noise, missing details, etc. From jamming sources. It requires a large amount of data and vast computational capabilities, an expansion of artificial intelligence to reach logical thinking, and it is in the program itself; it is very much like the mind of a small child is incomplete, but his flexibility is limitless. Sub-techniques have also been used; deep intervertebral decoders and deep neural networks.
Spread
The publication of the research findings at the International Symposium on the Implementation of the Digital Industry and Digital Transformation Management 2019, an academic conference on Artificial Intelligence, Robotics and the Techniques of the Fourth Industrial Revolution.
Wrong diagnosis
The work of ECG signals from these devices impedes the noise generated by the movement of the electrodes. This noise may cause a wrong diagnosis when a specialist is consulted. These signals are usually filtered using traditional noise cancellation techniques at specific frequencies.