Energy-efficient embedded machine learning

Description

Today, cardiovascular diseases represent one of the leading causes of death worldwide. Annual costs for treating people suffering from these cardiovascular diseases have been estimated at over $35 billion. In particular, there are three pathologies that are extremely prevalent:

• Atrial fibrillation (AF): AF is one of the most common types of cardiac arrhythmia, affecting more than 3 million people in the USA. This pathology progresses from the paroxysmal form to a persistent and/or a permanent one. AF is characterized by inter-patient variability. In particular, some patients do not experience any symptoms and the treatment is usually patient-dependant. Despite the progress in detection and treatment of AF, the arrhythmia remains one of the major risk factors for stroke and heart failure. In order to prevent progression of this pathology, there is a need for a personalized detection during the paroxysmal phase.

• Congestive Heart Failure (CHRF): Currently, there are nearly 5 million Americans suffering from CHRF. CHRF is a condition in which the heart is losing the capacity to pump enough blood to meet the body’s demand for oxygen and nutrients. It is usually preceded by an increase of fluid in the thoracic cavity, swelling of limbs, shortness of breath, quick weight gain as well as the irregular heartbeat. CHRF, if untreated, can cause damage to other important organs in the human body.

• Myocardial Infarction (MI): More than 700,000 people are affected by MI annually in the USA alone. MI, also commonly known as heart attack, occurs when one of the coronary arteries that supply the oxygenated blood to the heart muscle becomes blocked. This situation occurs due to a build-up of fatty deposits (plaques) that gradually form in one of these arteries. Upon rupture, these plaques release thrombogenic contents that trigger the blood clot to form. The blood clot can completely block an artery resulting in myocardial ischaemia, a diminished blood supply to the part of the heart that was getting supplied by the blocked artery. Without oxygen, muscle cells of this part of the heart begin dying, resulting in a heart attack.

Common practice so far has focused on monitoring the patients in hospitals for detection and prediction of cardiovascular pathologies. The existing solutions for monitoring cardiovascular pathologies are bulky, time-consuming, expensive, and intrusive. As a direct consequence, based on such solutions, it is not possible to monitor the patients on a long-term basis and in real time.

Our Approach

Wearable technologies offer a promising solution to pervasive healthcare at an affordable price, by removing the constraints with respect to time and location. More importantly, using wearable technologies, it is possible to monitor the cardiac functions of the patients in real time and on a long-term basis. This allows clinicians to detect early symptoms of potential cardiac irregularities and prevent further patient’s state deterioration. Subsequently, based on this analysis, care can be provided to patients, which reduces the hospitalization rate and healthcare costs. In this project, we monitor the cardiovascular function using the INYU sensor, a wearable electrocardiogram (ECG) monitoring device designed within the collaboration framework between the Embedded Systems Laboratory of EPFL and SmartCardia SA, which is shown below.

 

This project aims at early detection and prediction of cardiovascular pathologies in real time, through monitoring the ECG signal acquired by the INYU wearable sensor. For instance, AF episodes are associated with the absence of P-waves and irregular heart rate (HR) and MI is associated with a change in the morphology of heart beats, e.g., ST-segment abnormality. However, distinguishing each cardiovascular disease from other types of cardiovascular pathologies in the presence of noise still remains a challenging task. In order to tackle this problem, we have developed a machine learning method for cardiac-pathology classification based on the morphology of the ECG signal. This algorithm classifies whether a short ECG recording shows normal sinus rhythm (NSR), atrial fibrillation (AF), noisy signals (Noise), or alternative rhythm (OthR). The overall flow of our method is shown here and it consists of three main phases: pre-processing, feature extraction, and hierarchical classifier. The hierarchical classifier contains a multiclass classifier based on error-correcting output codes (ECOC) and a random forest classifier for binary decision making.

Related Publications

Real-Time Classification Technique for Early Detection and Prevention of Myocardial Infarction on Wearable Devices
D. Sopic, A. Aminifar, D. Atienza Alonso
2017	IEEE Biomedical Circuits and Systems Conference, Italy
A Patient-Specific Methodology for Prediction of Paroxysmal Atrial Fibrillation Onset
E. De Giovanni, A. Aminifar, A. Luca, S. Yazdani, JM. Vesin, D. Atienza Alonso
2017	Computing in Cardiology (CinC), France
Hierarchical Cardiac-Rhythm Classification Based on Electrocardiogram Morphology
D. Sopic, E. De Giovanni, A. Aminifar and D. Atienza Alonso
2017	Computing in Cardiology (CinC), France
Touch-Based System for Beat-to-Beat Impedance Cardiogram Acquisition and Hemodynamic Parameters Estimation
D. Sopic, S. Murali, F. J. Rincon Vallejos, D. Atienza Alonso
2016	Design Automation and Test in Europe Conference (DATE), Germany
Automated real-time atrial fibrillation detection on a wearable wireless sensor platform
F. Rincón, P.R. Grassi, N. Khaled, D. Atienza, D. Sciuto
2012	IEEE Engineering in Medicine and Biology Society Conference (EMBC)
A Self-Aware Epilepsy Monitoring System for Real-Time Epileptic Seizure Detection
Forooghifar, Farnaz ; Aminifar, Amir ; Cammoun, Leila ; Atienza Alonso, David ; Wisniewski, Ilona ; Ciumas, Carolina ; Ryvlin, Philippe
2019-08-08	Mobile Networks and Applications
Self-Aware Machine Learning for Multimodal Workload Monitoring During Manual Labor on Edge Wearable Sensors
Masinelli, Giulio ; Forooghifar, Farnaz ; Arza, Adriana ; Aminifar, Amir ; Atienza, David
2020-02-21	IEEE Design & Test
Resource-Aware Distributed Epilepsy Monitoring Using Self-Awareness From Edge to Cloud
Forooghifar, Farnaz ; Aminifar, Amir ; Atienza Alonso, David
2019-11-04	IEEE Transactions on Biomedical Circuits and Systems
Tailoring SVM Inference for Resource-Efficient ECG-Based Epilepsy Monitors
Atienza Alonso, David ; Ferretti, Lorenzo ; Ansaloni, Giovanni ; Pozzi, Laura ; Aminifar, Amir ; Cammoun, Leila ; Ryvlin, Philippe
2019	Design, Automation & Test in Europe Conference (DATE)
A wearable device for physical and emotional health monitoring
S. Murali, F. Rincon, D. Atienza
2015	Computing in Cardiology Conference (CinC)