User Tools

Site Tools


7a.021.TUT - Early Detection of Myocardial Infarction Using Echocardiogram Images

Project - Summary

The significant proportion (20-30%) of emergency department admissions are related to patients with acute chest pain. These patients are required to have a rapid assessment due to their timecritical condition. It has been shown that parameters such as changes in ECG characteristics or elevation of troponin level may detect only 30% of acute ischemic events. Here, echocardiography can play a valuable role as an alternative diagnostic tool in an appropriate triage of patients with acute chest pain. Echocardiography is a reliable method for revealing the anomalies in the regional heart wall motion. Due to the early manifestation of Myocardial Infarction (MI) symptoms in echocardiogram, this imaging modality is now included in the universal definition of acute MI and in international guidelines regarding the management of cardiac arrest. In this project, the ultimate goal is to design an automatic model, which traces the movement of the heart’s wall using the echocardiogram images and detects the anomalies in the wall motion. The initial focus of this work is detecting the Left ventricle (LV) muscle in each echo frames.

In this project, we evaluate our methods using a benchmark dataset including 152 echocardiogram videos from healthy and patients with MI. We first provide a pseudo labeling process to generate ground truth mask for LV location in each frame. Then, we used two state-of-the-art Convolutional Neural Networks (CNN) to segment the LV muscle. To the best of our knowledge, this is the first application of CNN for labeling the complete LV muscle at pixel-level. Numerical experiments demonstrate that both of the models have a robust and reliable performance for LV detection. Moreover, we design a flickering detection to be able to choose the minimum flickering among the predictions of different CNNs. Flickering evaluation is an essential step because it can easily lead to MI misclassification. Currently, we are waiting for MI labels from medical doctors. We shall continue this project in the next year to complete our end-to-end automatic MI detection.

Project - Team

Team Member Role Email Phone Number Academic Site/IAB
Moncef Gabbouj PI +358 (400) 736613 Tampere University
Serkan Kiranyaz Co-PI Not available Tampere University
Morteza Zabihi Student/Researcher Not available Not available Tampere University
Aysen Degerli Research Assistant/Student 358 46 521 9737 Tampere University
Sponsor: Tieto

Project - Novelty of Approach

To the best of our knowledge, this is the first study to analyze the single-lead ECG along with PCG signals. This cannot only increase the accuracy of anomaly detection but also encompasses a wider range of cardiac anomalies. In addition, the handheld devices can increase the ease of access and improve the accuracy of early anomaly detection.

(The proposed project depends on accessing a suitable dataset)

Project - Deliverables

1 (External) Dataset access and preparation
2 Semi-supervised segmentation (using anchor points)
3 Fully automatic segmentation (supervised learning)
4 Developing an end-to-end model for myocardial infarction detection

Project - Benefits to IAB

1. The proposed pseudo labeling process can be used for any other dataset with only a few annotated data. The constraint can be adapted for the application of interest.

2. To the best of our knowledge, this study provides the annotations for the first and largest MI echo dataset available in this domain.

3. The trained models in this study can be used for other dataset using transfer learning.

4. The provided LV masks can be used for further studies on the anatomical structure of heart muscle.

5. The proposed method can be used along with echocardiogram devices to provide a fast and reliable LV segmentation for cardiologists. This information can lead to improvement of the MI diagnosis.

Project - Presentation Video

Project - Documents

FilenameFilesizeLast modified
7a.021.tut_ip_info_sheet.docx140.6 KiB2019/08/20 11:47
7a.021.tut_final_report.pdf984.6 KiB2019/08/20 09:14
7a.021.tut_confluence_project_page.pdf238.4 KiB2019/08/13 15:03
7a.021.tut_quad_chart_2018_spring_meeting.pptx533.5 KiB2019/08/13 15:03
7a.021.tut-ecocardiogram-executive-summary.pdf525.7 KiB2019/08/13 15:03
7a.021.tut_personalized_heart_assessment_quad_2017_fall_meeting.pptx1.5 MiB2019/08/13 15:03
7a.021.tut_cvdi-mid-year-report-zabihi.pdf443.9 KiB2019/08/13 15:03
projects/year7/7a.021.tut.txt · Last modified: 2021/06/02 15:06 by sally.johnson