應用於心肌梗塞偵測之硬體加速器設計

Design of Hardware Accelerator for Myocardial Infarction Detection

簡暉紘(Huei-Hong Jian)

指導教授 ： 施鴻源

淡江大學/工學院/電機工程學系碩士班/碩士(2022年)

https://doi.org/10.6846/TKU.2022.00167

摘要

心臟相關疾病是全球性的公共衛生問題，但判讀心電圖上是一項浩大的工程，因此國內外開始引進AI做快速篩檢，不過神經網路往往需要計算大量數據，因此花費大量的運算時間與資源，為解決此問題，本論文透過軟硬協同設計(hardware software co-design)利用現場可編成邏輯門陣列(Field Programmable Gate Array, FPGA)來實現應用於偵測心肌梗塞疾病的硬體加速器。本論文主要分為四個部分來實現: (1)利用PhysioNet中的PTB Diagnostic資料庫為原始訊號，透過MATLAB進行資料預處理。(2)利用Python進行辨識心肌梗塞疾病之訓練，並將模型中的權重與偏差值進行資料處理。 (3)利用NIOS II將處理好的訓練資料、權重值、偏差值透過Avalon Bus匯流排寫入FPGA內的SDRAM中(4)利用Avalon Bus匯流排將SDRAM中的資料載入硬體架構中進行運算。透過驗證可得知本論文提出的架構可以有效的偵測心肌梗測病徵，與Python相互對比得知，本文所提出設計架構具有更快的運算效能且準確度達到95%以上。

關鍵字

心電圖 ； 神經網路 ； 定點數 ； 軟硬協同 ； 硬體加速器

並列摘要

Heart-related diseases are a global public health problem, but interpreting ECG is a huge project, since that AI has be introduced for rapid screening. However, neural networks often need to calculate a large amount of data, so it takes a lot of computing time and resources. In order to solve this problem, this paper using FPGA to implement a hardware accelerator for detecting myocardial infarction disease through hardware software co-design. This paper used four parts to realize the whole structure. (1) Using the PTB Diagnostic database in PhysioNet as the original signal, and preprocessing data through MATLAB. (2) Using Python to train the identification of myocardial infarction disease, and processing the weights and biases in the model. (3) Using NIOS II to write the processed input data, weights, and biases into the SDRAM in the FPGA through the Avalon Bus (4) Using the Avalon Bus to load the data in the SDRAM into the hardware architecture for operation. Through the verification, the architecture in this paper can effectively detect the myocardial infarction disease. Compared with Python, this paper proposed architecture can increase performance and has 98% accuracy.

並列關鍵字

ANNs ； FPGA ； fixed points ； hardware accelerator ； ECG ； hardware software co-design

參考文獻

[1] https://dep.mohw.gov.tw/dos/lp-5069-113-xCat-y109.html

[2] https://www.nhi.gov.tw/Content_List.aspx?n=D529CAC4D8F8E77B topn=23C660CAACAA159D

[3] Bousseljot R, Kreiseler D, Schnabel, A. Nutzung der EKG-Signaldatenbank CARDIODAT der PTB über das Internet. Biomedizinische Technik, Band 40, Ergänzungsband 1 (1995) S 317

[4] PhysioNet, PhysioNet ATM,URL：https://archive.physionet.org/cgi-bin/atm/ATM

[5] Terasic, Altera 多媒體發展平台 DE2-115,