Artificial intelligence (AI) systems have shown great promise in diagnosing and predicting different diseases and analyzing medical images. They can perform certain tasks better than human doctors, like surgical stitching and diagnosing autism in infants.

According to a report on IEEE Spectrum’s biomedical engineering blog, researchers at the University of Nottingham in the UK created a system that scanned patients’ routine medical data and predicted who of them will suffer heart attacks or brain attacks (strokes) within 10 years. Compared to the standard prediction method, the AI system correctly predicted the fates of 355 more patients [1].

Currently, the main standard method of assessing a patient’s risk is based on guidelines developed by the American Heart Association and American College of Cardiology. Doctors use these guidelines, focusing on well-established risk factors like high blood pressure, cholesterol, age, smoking, and diabetes, to develop counsel and treatment for their patients.

The researchers used clinical records and fed them into their machine learning models to find the distinguishing characteristics of patients who experience heart attacks or strokes.

According to statistic where a score of 1.0 signifies 100 percent accuracy, the standard methods got a score of 0.728. The models of machine learning ranged from 0.745 to 0.764, with the best score coming from a type of machine learning model called a neural network.

Together with Dr. Gillian Pearce, former senior lecturer in Biomedical Sciences at the University of Wolverhampton in the UK, I undertook the task of developing an AI that can diagnose stroke itself.

monitoring brain attack artificial intelligence stroke

Fig. 1. Remote Patient Monitoring

An AI capable of doing so would be able to expand the availability of emergency stroke treatment nearly 60-fold, bringing life-saving treatment to poor and rural people around the world.

We had previously prototyped just such an AI and now we have developed a new artificial neural network (ANN) model to endow it with the ability to communicate directly with the patient and refined the AI into a clinically-useful product that can be used in small hospitals and ambulances worldwide.

An ANN is a mathematical representation of the human neural architecture, reflecting its “learning” and “generalization” abilities. For this reason, ANNs belong to the field of AI.

Artificial neural network models [2 -7, 8-10] and statistical models were developed [11, 12], genetic algorithms and other machine learning techniques were used including decision tree, fuzzy sets, and evolutionary algorithms in medicine for different diseases diagnosis. The models used on practical study of patients, for the early diagnosis of myocardial infarction, chest pain, heart attack.

The model discussed in this article is a neural network model with 16 inputs which are a combination of symptoms and the risk factors of stroke provided by the patients. The presence of symptom and risk factor is 1 and absence is 0. In this model two hidden layers have been used. Output layer consists of one node which represents the probability of occurrence of stroke.

Our experience in this realm of AI is unique. To our knowledge, no one else has ever attempted to develop an AI capable of diagnosing stroke patients. We have not only attempted this feat, but the model has succeeded: our prototype AI distinguishes between stroke patients and normal subjects with > 99% accuracy.

monitoring brain attack artificial intelligence stroke

Dr. Lela Mirtskulava presenting at AIMed 2017

We hope that our AI diagnosis can directly benefit the millions of people who suffer stroke around the world each year.

According to the Centers for Disease Control and Prevention, 1,700 people die from heart disease every day where the main reason is the absence of qualified and immediate assistance [13]. According to the World Health Organization, 17 million people suffer stroke each year worldwide. Of these, 6.5 million die.

High blood pressure contributes to more than 12.7 million strokes worldwide. Europe averages approximately 650,000 stroke deaths each year. In developed countries, the incidence of stroke is declining, largely due to efforts to lower blood pressure and reduce smoking. However, the overall rate of stroke remains high due to the aging of the population [14].

Stroke occurs when a blood vessel is clogged by a blood clot and blood supply to that part of the brain is cut off. Brain cells will be damaged or die without blood. Depending on how quickly the person is treated, the effects of stroke on survivors can be devastating to a person’s body, mobility and speech, as well as how they think and feel.

Stroke is the main cause of death and disability worldwide. It can happen to anyone and at any age, and impacts everyone including survivors, family and friends, and communities. From making individual or global changes we can all do something to prevent stroke.

Permanent health monitoring will save up to 70% of human lives through timely diagnosis to prevent heart or brain attacks, when every minute counts.

Strokes that happen during the daytime are usually witnessed by the people around the patient. Therefore, the stroke onset time is more likely to be known and precisely recorded. However, a stroke can occur during sleep at night and the time at which the stroke happened is less likely to be known.

It is very important to know what the precise time of the stroke onset is, since in most cases strokes are suitable for treatment with Alteplase (a clot busting agent and invasive method for stroke patient treatment).

monitoring brain attack artificial intelligence stroke

Fig. 2. Mobile Patient Dashboard – Clinical Data

However, Alteplase must be administered within 4.5 hours of the stroke happening to be effective. Unfortunately, if a stroke arises while the patient is asleep, it is difficult to know the stroke onset time.

It is also known that if a stroke happens, the limbs (arms, legs) may be affected by the stroke causing motor function loss. The affected limbs would not move independently, as they usually do during sleep.

Therefore, if we had some technique for monitoring the normal movement of an individual’s limbs during the night, by recording any profound changes in normal movement, we would have a system of knowing the time of the stroke occurrence [15].

Additionally, the system could alert a hospital or relatives. This would enable immediate transfer of the stroke patient to hospital and timely administration of Alteplase if needed.

We are developing such a system based on internet of things (IoT) and neural network application equipped with a portable Bluetooth ECG wearable monitor.

Sensor Data obtained from the application have been collected by using different types of sensors which can pass the data (offline or in real time) to the ANN for analysis.

A real-time QRS Detection Algorithm can detect possible problems and automatically sends 1 Hour of electrocardiogram (ECG) raw data. The patient can also trigger the data send. An Android phone can act as the communicator.

Smartphones have been recognized as a possible tool for telemedicine since they became commercially available. They can interact with electromedic devices (EMDs), like patient monitors, and transmit vital signals over internet protocols, such as TCP/IP and UDP.

Furthermore, the neural network itself can be trained to recognize an individual patient’s normal movement and therefore to recognize limb motion weakness (as occurring in a stroke).

As a result, we have developed an ANN for medical diagnosis. This study aimed to evaluate an ANN in stroke diagnosis. The feedforward back propagation neural network with supervised learning was proposed to diagnose stroke.

The ANN provides a powerful tool to help medical staff to analyze, model and make sense of complex clinical data across a broad range of medical applications. ANNs showed significant results in dealing with data represented in symptoms.

ANNs with the ability to learn by example are a very flexible and powerful tool in medical diagnosis, offering very useful applications to modern medicine.



  1. os/biomedical/diagnostics/ai-predicts- heart-attacks- more-accurately-than- standard-doctor- method
  1. Catalogna M, Cohen E, Fishman S, Halpern Z, Nevo U, Ben-Jacob E. Artificial neural networks based controller for glucose monitoring during clamp test. PloS One. 7: e44587, 2012.
  1. Elveren E, Yumuşak N. Tuberculosis disease diagnosis using artificial neural network trained with genetic algorithm. J Med Syst. 35: 329–332, 2011.
  1. Dey P, Lamba A, Kumari S, Marwaha N. Application of an artificial neural network in the prognosis of chronic myeloid leukemia. Anal Quant Cytol Histol. 33: 335–339, 2012.
  1. Barwad A, Dey P, Susheilia S. Artificial neural network in diagnosis of metastatic carcinoma in effusion cytology. Cytometry B Clyn Cytom. 82: 107–111, 2012.
  1. Barbosa D, Roupar D, Ramos J, Tavares A and Lima C. Automatic small bowel tumor diagnosis by using multi-scale wavelet-based analysis in wireless capsule endoscopy images. Biomed Eng Online. 11: 3, 2012.
  1. T. Zrimec and I. Kononenko, “Feasibility analysis of machine learning in medical diagnosis from aura images”, University of Ljubljana, 2004.
  1. Og˘uz Karan, Canan Bayraktar, Haluk Gumuskaya, Bekir Karlık.Diagnosing diabetes using neural networks on small mobile devices.Elsevier. Expert Systems with Applications 39 (2012) 54–60.
  1. S. Moein, S.A. Monadjemi, and P. Moallem, “A Novel Fuzzy Neural Based Medical Diagnosis system”, International Journal of biological and Life Science 4:3, 2008.
  1. Irfan Y. Khan, P.H. Zope, S.R. Suralkar. Importance of Artificial Neural Network in Medical Diagnosis disease like acute nephritis disease and heart disease. International Journal of Engineering Science and Innovative Technology (IJESIT) Volume 2, Issue 2, March 2013
  1. K. Hung, and Y.T. Zhang, “Implementation of a WAP-Based Telemedicine System for Patient Monitoring”, IEEE Transactions on Information Technology in Biomedicine, vol. 7, nº 2, June, 2003.
  1. Salah M. Saleh Al-Majeed, Shavan K. Askar, and Martin Fleury. H.265 Codec over 4G Networks
  2. reasons-build- health-monitoring- system-hospital
  3. Lela Mirtskhulava, Julian Wong, Gillian Pearce, Salah Al-Majeed. Artificial Neural Network Model in Stroke Diagnosis. DOI: 10.1109/UKSim.2015.33 Conference: 2015 17th UKSIM-AMSS International Conference on Modelling and Simulation, At Cambridge, UK
  1. Gillian Pearce, Lela Mirtskhulava, Julian Wong, Salah Al-Majeed, Koba Bakuria, Nana Gulua. DOI: 10.1109/UKSim.2015.34 Conference: 2015 17th UKSIM-AMSS International Conference on Modelling and Simulation, At Cambridge, UK

BIOmonitoring brain attack artificial intelligence stroke

Dr. Lela Mirtskulava currently holds a position of Senior Scientist at NeuroSpring, USA and Associate Professor at the Department of Computer Science, Faculty of Exact and Natural Sciences at Ivane Javakhishvili Tbilisi State University. She is a faculty member of San Diego State University Georgia. She also held the position of Program Coordinator/Quality Assurance Manager/Associate Professor and a Head of Research Center for Mobile Computing at the Faculty of Computer Technologies and Engineering at International Black Sea University, Associate Professor at Sokhumi State University and Georgian Technical University in past. She also has 13 years of industry work experience as an ICT Senior Engineer at Geocell LLC, Georgia. Her research interests include: Artificial Intelligence diagnostic for Stroke treatment, human activity recognition, power consumption, energy efficiency, mobile app development, software development, computer networks and security, bioinformatics, wireless communications, climate protection and global warming. Dr. Mirtskhulava’s experience in this realm of artificial intelligence is unique. No one else has ever attempted to develop an artificial intelligence capable of diagnosing stroke patients. Dr. Mirtskhulava has not only attempted this feat but she has succeeded: her prototype artificial intelligence distinguishes between stroke patients and normal subjects with > 99% accuracy. Dr. Mirtskhulava has received her PhD Degree in Computers, Complexes, Systems and Networks at the Department of Informatics at the Georgian Technical University in 1998. She is the recipient of DAAD Scholarship Certificate in scope of Academic staff exchange program, at Westsaxson University of Applied Sciences Zwickau, Germany in 2016. Dr. Mirtskhulava has participated in Professor Exchange program to San Diego State University, CA, USA in 2016. She was a Visiting Professor at University of Cambridge, UK in 2013. She has published over 43 refereed scientific papers, many of which appeared in IEEE and ACM digital libraries and International conferences such as AIMed, UKSim, BlackSeaCom, WCECS, She has received Best Paper Award and Session Chair Certificate at UKSim 2015 in Cambridge, UK. She was a Speaker at the International conference “Artificial Intelligence in Medicine” AIMed 2017 in Los Angeles, the First SDSU – Georgia STEM WORKSHOP on Nanotechnology and Environmental Sciences in 2015.