Detection of Abnormalities in Cardiac Rhythm Using Spiking Neural Networks

Abstract

Artificial Intelligence (AI) and Machine Learning (ML) have been increasingly attracting attentions in many different fields. Healthcare is one of the areas that
has greatly benefited from the advances in AI/ML. This includes a wide range of applications such as medical data interpretation, disease or abnormality detection
or prediction, monitoring specific health condition and medical data management. On the other hand, patients can also take advantage of available healthcare devices
to be more conscious of their health status and increase their quality of life. However, implementing AI/ML algorithms on resource-constrained wearable
devices is challenging. One way to tackle this problem is to exploit the neuromorphic computing solutions such as Spiking Neural Networks (SNNs), which are
more energy efficient than conventional neural networks because of their more similar function to how the brain works. In this thesis project, we investigate the
working-mechanism of these networks, how we can design, train and use them for the detection of abnormalities in cardiac function.

Popular Abstract

Artificial Intelligence (AI) and Machine Learning (ML) are terms being heard more and more these days. The reason is that the AI/ML based methods are showing promising performance in many areas and are still finding their ways into broader range of applications and the speed of advances in this field is astonishing. Healthcare is one of the areas that has significantly benefited from AI/ML advances. This includes a wide range of applications such as medical data interpretation, disease or abnormality detection or prediction, monitoring specific health condition and medical data management.
On the other hand, nowadays, many people have the experience of using wearable devices such as smart watches. These devices can be used to keep track of different kinds of exercise activities, provide information about the user’s sleep cycles, extract several different individual patterns during the day, detect an abnormality in the body function and warn the user about it. These applications have been made possible with the help of incorporated sensors to collect data and implemented algorithms to use the data and extract valuable information. These algorithms are mostly based on AI/ML. The healthcare wearables can help people to be more conscious of their health status and increase their quality of life.
However, one of the main challenges is that these devices are usually small and resource-constrained and high performance algorithms such as state of the art neural networks are not a practical option to use. An idea is to get more inspiration from the brain and the function of biological neurons, and the intuition behind it is that our brain is significantly energy efficient and operates on about 20 watts. One of such brain inspired solutions is using Spiking Neural Networks (SNNs), which are more energy efficient than conventional neural networks.
In this thesis project, we investigate the working-mechanism of these networks, how we can design, train and use them for the detection of abnormalities in cardiac function.

• author: Mohammad, Dorsa
• supervisor: Amir Aminifar
• organization: Department of Electrical and Information Technology
• course: EITM02 20221
• year: 2023
• type: H2 - Master's Degree (Two Years)
• subject: Technology and Engineering
• keywords: Artificial Intelligence (AI), Machine Learning (ML), Spiking Neural Network (SNN), Artificial Neural Networks, Healthcare devices, Electrocardiogram (ECG), Neuromorphic computing.
• report number: LU/LTH-EIT 2023-915
• language: English
• additional info: The host company of this project is Ericsson. The supervisor from Ericsson is Dr. Saeed Bastani.
• id: 9113248
• date added to LUP: 2023-04-13 10:43:12
• date last changed: 2023-04-13 10:43:12