Gerçek zamanlı gömülü sistemler için rastgele orman tabanlı şok tavsiye algoritması

Abstract

Ani kalp durması (SCA), kalbin kan dolaşımını sağlayamayacak kadar etkisiz hale gelmesi durumudur. SCA'ya sebeb olan en yaygın ritim bozukluklarından ikisi ventriküler fibrilasyon ve ventriküler taşikardidir. Bu ritim bozukluklarının düzeltilmesi için kalbe elektrik şoku verilmesi hayati önem taşır. Otomatik Harici Defibrilatör (AED) cihazları, hastanın kalp ritmini analiz eder ve gerektiğinde otomatik olarak elektrik şoku uygular. AED cihazı, ritim bozukluklarının tespiti için elektrokardiyogram sinyallerini toplar, şok tavsiyesi algoritmaları (SAA) ile hastaya şok verilip verilmeyeceğine karar verir. Ancak AED’de yürütülecek bir SAA geliştirme süreci, gömülü sistemlerin sınırlı veri işleme kapasitesi gibi bazı zorluklar barındırmaktadır. Algoritma gömülü sistemde çalışacağı için kısıtlı veri işleme kabiliyeti olan bir ortamda kalp ritimlerini başarıyla ayırt edebilecek nitelikte olmalıdır. Mevcut çalışmalar incelendiğinde, eşik tabanlı ya da makine öğrenmesi (ML) temelli SAA’ların bulunduğu ve birçok ML tabanlı SAA’nın yüksek sınıflandırma başarısı sunduğu görülmektedir. Ancak, bu algoritmaların yalnızca küçük bir kısmı gerçek zamanlı gömülü sistemlerde test edilmiş ve AED cihazlarına uygunluğu değerlendirilmiştir. Bu çalışmada, halka açık bir veri seti kullanılarak, geleneksel bir ML yöntemi olan rastgele orman ile SAA geliştirilmiştir. Önerilen algoritma, şok uygulanabilir ritimlerde %92.9 duyarlılık ve şok uygulanmaması gereken ritimlerde %99.2 özgüllük değerleri sağlamıştır. Yüksek seviye programlama dili ile geliştirilen SAA, C diline entegre edilerek mikrodenetleyici tabanlı bir geliştirme kitinde test edilmiştir. 500 kB bellek ihtiyacı ve 75 mikro saniyelik tespit süresi ile algoritmanın AED cihazlarında kullanım için uygun olduğu ve başarıyla entegre edilebileceği kanıtlanmıştır.

Keywords

• AED
• Şok Tavsiyesi Algoritması
• Makine Öğrenmesi

A random forest-based shock advice algorithm for real-time embedded systems

Abstract

Sudden cardiac arrest (SCA) occurs when the heart becomes unable to pump blood effectively. Two of the most common arrhythmias that cause SCA are ventricular fibrillation and ventricular tachycardia. Treating these arrhythmias through defibrillation—delivering an electric shock to the heart—is vital for patient survival. Automated External Defibrillators (AEDs) analyze the patient’s heart rhythm and automatically deliver a shock when necessary. To do this, AEDs collect electrocardiogram (ECG) signals and use shock advisory algorithms (SAAs) to decide whether a shock is required. However, the development of SAAs suitable for AEDs involves challenges, such as the limited data processing capacity of embedded systems. The algorithm must be capable of reliably distinguishing heart rhythms in such constrained environments. A review of existing studies reveals both threshold-based and machine learning (ML)-based SAAs, with many ML-based algorithms demonstrating high classification performance. Yet, only a small fraction of these algorithms have been tested in real-time embedded systems, and their applicability to AED devices has been evaluated in limited contexts. In this study, a traditional ML-based SAA was developed using the random forest method and evaluated with a publicly available dataset. The proposed algorithm achieved 92.9% sensitivity for shockable rhythms and 99.2% specificity for non-shockable rhythms. Initially developed in a high-level programming language, the SAA was integrated into C and tested on a microcontroller-based development kit. With a memory requirement of 500 kB and a detection time of 75 microseconds, the algorithm was shown to be suitable for implementation in AED devices, demonstrating its potential for commercial use.

Keywords

• AED
• Shock Advice Algorithm
• Machine Learning

References

1. [1] American Heart Association. “Heart Attack and Sudden Cardiac Arrest Differences”. https://www.heart.org/en/health-topics/heart-attack/about-heart-attacks/heart-attack-or-sudden-cardiac-arrest-how-are-they-different (10.11.2024).
2. [2] American Heart Association. “Arrhythmias”. https://watchlearnlive.heart.org/index.php?moduleSelect=arrhyt (10.11.2024).
3. [3] Congress.gov. "H. Rept. 118-520 - Cardıomyopathy Health Educatıon, Awareness, And Research, And Aed Traınıng In The Schools Act Of 2024". https://www.congress.gov/congressional-report/118th-congress/house-report/520/1 (10.11.2024).
4. [4] Acharya R, Fujita H, Oh S, Raghavendra U, Tan J, Adam M, Gertych A, Hagiwara Y. “Automated identification of shockable and non-shockable life-threatening ventricular arrhythmias using convolutional neural network”. Future Generation Computer Systems, 79, 952–959, 2018.
5. [5] Nguyen M, Nguyen T, Le H. “A review of progress and an advanced method for shock advice algorithms in automated external defibrillators”. BioMedical Engineering Online, 21, 2022.
6. [6] Gurkan H, Hanilci A. “ECG based biometric identification method using QRS images and convolutional neural network”. Pamukkale University Journal of Engineering Sciences, 26, 2, 318–327, 2020.
7. [7] Figuera C, Irusta U, Morgado E, Aramendi E, Ayala U, Wik L, Kramer-Johansen J, Eftestøl T, Alonso-Atienza F. “Machine learning techniques for the detection of shockable rhythms in automated external defibrillators”. PLoS One, 11, 7, 2016.
8. [8] Nasimi F, Yazdchi M. “LDIAED: A lightweight deep learning algorithm implementable on automated external defibrillators”. PLoS One, 17, 2, 2022.

9. [9] Khadar S, Tabatabaey-Mashadi N, Daliri G. “A simple realtime algorithm for automatic external defibrillator”. Biomedical Signal Processing and Control, 51, 277–284, 2019.
10. [10] Orozco-Duque A, Rúa S, Zuluaga S, Redondo A, Restrepo J, Bustamante J. “Support vector machine and artificial neural network implementation in embedded systems for real time arrhythmias detection”. BIOSIGNALS 2013-Proceedings of the International Conference on Bio-Inspired Systems and Signal Processing, 310–313, 2013.
11. [11] American Heart Association. “What is an Arrhythmia”. https://www.heart.org/en/health-topics/arrhythmia/about-arrhythmia (10.11.2024).
12. [12] Krasteva V, Ménétré S, Didon J, Jekova I. “Fully convolutional deep neural networks with optimized hyperparameters for detection of shockable and non-shockable rhythms”. Sensors (Switzerland), 20, 10, 2020.
13. [13] Moody G, Mark R. “The impact of the MIT-BIH arrhythmia database”. IEEE Engineering in Medicine and Biology Magazine, 20, 3, 45–50, 2001.
14. [14] Moody G, Mark R. “MIT-BIH Arrhythmia Database”. https://physionet.org/content/mitdb/1.0.0/ (10.11.2024).
15. [15] Tripathy R, Sharma L, Dandapat S. “Detection of Shockable Ventricular Arrhythmia using Variational Mode Decomposition”. Journal of Medical Systems, 40, 4, 1–13, 2016.
16. [16] Nam D, Kang D, Myoung H, Lee K. “Detection method for shockable rhythm based on a single feature”. Electronics Letters, 52, 9, 686–688, 2016.
17. [17] Nishiyama T, Nishiyama A, Negishi M, Kashimura S, Katsumata Y, Kimura T, Nishiyama N, Tanimoto Y, Aizawa Y, Mitamura H, Fukuda K, Takatsuki S. “Diagnostic accuracy of commercially available automated external defibrillators”. Journal of the American Heart Association, 4, 12, 2015.
18. [18] STMicroelectronics, “STM32 Nucleo-144 development board with STM32F429ZI MCU”. https://www.st.com/en/evaluation-tools/nucleo-f429zi.html (10.11.2024).
19. [19] Greenwald S. “The MIT-BIH Malignant Ventricular Arrhythmia Database”. https://archive.physionet.org/physiobank/database/vfdb/ (10.11.2024).
20. [20] Moody G. “MIT-BIH Normal Sinus Rhythm Database”. https://physionet.org/content/nsrdb/1.0.0/ (10.11.2024).