Presentation Information
[18a-C31-9]Quantum Stochastic Resonance-Based Reservoir Computing System for Hypertensive and Diabetic MCG Diagnosis
〇Xiaoyu Shi1, Zhiqiang Liao1, Hitoshi Tabata1 (1.Tokyo Univ.)
Keywords:
bioengineering,nonlinear physics,reservior computing
Cardiovascular diseases (CVDs) have become the biggest threat to human health, and they are accelerated by hypertension and diabetes. The best way to avoid the many complications of CVDs is to manage and prevent hypertension and diabetes at an early stage. Thus, developing an accurate and automatic diagnosis method becomes important. Magnetocardiography (MCG) is a noninvasive contactless method to measure the magnetic field generated by the same ionic currents that create the electrocardiogram (ECG). However, compared with ECG, 3-dimentional MCG is a faster and contactless method with higher spatial and temporal resolution. Therefore, this study explores the applicability of three-axes MCG in hypertension and diabetes diagnosis.
In order to reach the goals of accuracy and high speed at the same time, I proposed a machine learning method for automatic hypertension and diabetes MCG diagnosis. The method is based on a system using quantum stochastic resonance-based reservoir computing system (QSRRC). This system innovatively leverages the strong memory and nonlinearity characteristics of QSRRC to memorize and learn from hypertension and diabetes features. There are mainly three steps: (1) 3-dimentional MCG data preprocessing, (2) feature extraction, and (3) classification.
Instead of human, we chose to apply the proposed method on rats’ MCG dataset first. We measured ten each of hypertensive (SHR/lzm), type 1 diabetic (KDP/Slc), and healthy (Slc:SD) 8-week-old male rats with optically pumped magnetometers (OPM) inside a shielded box. When we used the MCG signals of x, y, and z axes separately for diagnosis, the average diagnostic accuracies reached 92.50%, 95.96%, and 94.73%, respectively, which were on average 34% higher than that of the traditional RC system. Furthermore, the STM and PC tasks were benchmarked and compared on traditional RC system and QSRRC model, highlighting the memory effect of proposed system and explaining the better performance.
In order to reach the goals of accuracy and high speed at the same time, I proposed a machine learning method for automatic hypertension and diabetes MCG diagnosis. The method is based on a system using quantum stochastic resonance-based reservoir computing system (QSRRC). This system innovatively leverages the strong memory and nonlinearity characteristics of QSRRC to memorize and learn from hypertension and diabetes features. There are mainly three steps: (1) 3-dimentional MCG data preprocessing, (2) feature extraction, and (3) classification.
Instead of human, we chose to apply the proposed method on rats’ MCG dataset first. We measured ten each of hypertensive (SHR/lzm), type 1 diabetic (KDP/Slc), and healthy (Slc:SD) 8-week-old male rats with optically pumped magnetometers (OPM) inside a shielded box. When we used the MCG signals of x, y, and z axes separately for diagnosis, the average diagnostic accuracies reached 92.50%, 95.96%, and 94.73%, respectively, which were on average 34% higher than that of the traditional RC system. Furthermore, the STM and PC tasks were benchmarked and compared on traditional RC system and QSRRC model, highlighting the memory effect of proposed system and explaining the better performance.
Comment
To browse or post comments, you must log in.Log in