Cardiac Arrhythmias detection Using Artificial Neural Network

Cardiac Arrhythmia are an important class of Cardio-vascular diseases. Most arrhythmia are benign and have transitory nature but some are very serious and need timely diagnosis. Most often the arrhythmia episodes occur randomly, it therefore becomes very difficult to capture an episode using normal

Project Title

Cardiac Arrhythmias detection Using Artificial Neural Network

Project Area of Specialization

Artificial Intelligence

Project Summary

Cardiac Arrhythmia are an important class of Cardio-vascular diseases. Most arrhythmia are benign and have transitory nature but some are very serious and need timely diagnosis. Most often the arrhythmia episodes occur randomly, it therefore becomes very difficult to capture an episode using normal ECG recorder. A 24 hours ECG recording (Holter monitoring) system is therefore needed. However a life threatening episode monitored after 24 hours is of no use. It is therefore desirable to have a system that can instantly prompt about the occurrence of such events. The proposed system shall detect arrhythmic events in real time. The system makes use of machine learning algorithm running on a wearable Raspberry Pi computer.

To train the ML algorithm pre-labeled ECG QRS complexes are required. Prerecorded digitized ECG available at Physionet as MIT-BIH database is bandpass-filtered and preprocessed to separate the QRS complexes from the record. These complexes together with their labels showing the arrhythmia type, are applied for training the ML algorithm. An ECG acquisition system captures the real time ECG and digitizes it. The computer running the trained ML algorithm classifies the ECG into Normal Sinus Rhythm or one of the arrhythmia types. The system is equipped with GSM modem. If a dangerous arrhythmia is detected, the system shall transmit the full episode as a chunk of ECG complexes to the healthcare center for immediate action.

Project Objectives

1. Learning the Deep learning algorithms and programming one in Python language.
2. Learning about Cardiac Arrhythmia and programming a machine learning algorithm for their detection.
3. Learning about Raspberry Pi computer and programming it in Python language for ECG processing, arrhythmia detection and transmitting a few seconds ECG chunk through GSM modem.
4. Programming the Raspberry Pi for receiving ECG through GSM module and displaying it.
5. Interfacing an ECG acquisition system with Raspberry Pi.

Project Implementation Method

The project starts with a thorough literature survey. Next, the ECG data available at MIT-BIH database is downloaded, filtered and processed to separate the QRS complexes around R-peak as marked in the annotation files. The digital samples of these complexes and their arrhythmia types, as marked along each R-peak sample in the associated annotation file, form the input and target output respectively, for the ML algorithm. The data is separated into a training and a testing subset. The training data is used to train the network. A deep learning NN shall be used for this purpose. The performance of the trained network shall then be assessed with the test data. 

For use in real time a wearable system must be designed. The system worn by a patient would continuously acquire ECG and detect the type of beat. For this purpose an ECG acquisition system comprising of an amplifier and bandpass filter is used. The acquired data is digitized and fed to a Raspberry Pi computer. Coding is done to detect the R-peaks of the ECG in real time and applying the Pan-Tompkins algorithm to find the PQRST segments around each R-peak.  The PQRST segments are then fed to the trained Neural Network running on the same Raspberry Pi computer. The system detects the beat type. If it is a normal sinus beat, nothing happens but if a dangerous Arrhythmia is detected, the system transmits the full chunk of ECG to the healthcare center. A GSM modem interfaced to the Raspberry Pi is used for this purpose. On the receiving end a GSM modem interfaced to a PC receives the ECG and displays it along with the patient ID. In case of emergency the hospital takes the necessary action.

Benefits of the Project

The project is expected to play a key role in providing healthcare solution for a very important health problem. Constant monitoring of heart condition is very crucial for heart patients. Patients that have more serious heart condition need hospitalization but those who are otherwise fine but occasionally develop cardiac arrhythmia cannot be hospitalized and need remote monitoring. The system proposed in this project solves this purpose. On one hand the patient is constantly monitored while on the other hand he/she has the liberty to live a normal daily life. The system would also relieve the patient of frequent visits to the hospital especially in places where the patient’s residence is far away from hospital and commuting is difficult. This will also reduce the patient load on hospitals.

Technical Details of Final Deliverable

A code shall be written in Python language for processing the ECG downloaded from MIT-BIH database. After segmenting the QRS complexes the samples of these complexes and their associated target arrhythmia types are saved in an array. This data is fed to a Deep Learning Neural Network coded in Python. A trained network is thus created.

A wearable ECG acquisition system is also used, that consists of an ECG amplifier that acquires a single lead ECG of the patient and filters it to remove noise and artefacts and digitizes it.

A wearable Raspberry Pi computer running a code written in Python takes the digitized ECG as input and processes it to segment the QRS complexes from the ECG in real time. Each complex is fed as input to the trained Deep learning NN imported to this computer. On detection of an Arrhythmic beat the Raspberry Pi computer transmits the full chunk of ECG consisting of the arrhythmic beat, using a GSM modem.

A Raspberry Pi computer dedicated as the doctor’s console is programmed in Python language for receiving the GSM modem data through its interface and displaying it on the monitor.

Final Deliverable of the Project

1. Software code in Matlab for downloading ECG recordings from MIT-BIH database.
2. Software Code in Python language for processing the downloaded ECG, segmenting it and training a Deep learning NN.
3. Software code in Python for the wearable Raspberry Pi computer that is interfaced to the wearable ECG acquisition system. The code segments the ECG in real time and feeds the complexes to the imported Deep learning NN for arrhythmia detection. On detecting an arrhythmic beat the ECG chunk is transmitted through GSM modem
4. Software code in Python for Raspberry Pi computer used as doctor’s console for receiving and viewing the ECG.
5. A wearable ECG acquisition hardware. 

Final Deliverable of the Project

HW/SW integrated system

Core Industry

Medical

Other Industries

IT

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Good Health and Well-Being for People

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com