本文提出應用於孕婦腹部之胎兒心電訊號(electrocardiogram, ECG)識別與監測系統，使用深度學習(deep learning)之卷積神經網路(convolutional neural network)演算法來達到此目的。主要透過無侵入式(noninvasive)感測器電極量測出孕婦腹部的心電訊號，並透過無線傳輸再經由電腦分析出胎兒心電訊號，可以快速並準確的處理長時間的心電訊號數據流，達到監控胎兒健康狀況的目的。我們從孕婦腹部心電資料庫擷取心電訊號以250ms做樣本分割(segmentation)，接著設立前後共20ms的保護區間(protection interval)，確保樣本分割時心電波形的完整性，以方便正確標示(label R peak)心電訊號的類別。為了降低在時域上雜訊對心電圖所造成的影響，透過短時距傅立葉轉換(short-time Fourier transform)，將心電訊號由時域轉為頻域，使得心電訊號可以二維時頻的形式表現，再經由卷積神經網路訓練與測試數據。由於心電資料庫為使用四條導線量測紀錄孕婦腹部心電訊號，最後採用融合(fusion)的方法，對於不同電極之採用不同權重進行分類結果的加權，從而該心電訊號波型之類別。以編號r01、r08與r10的孕婦之個別偵測率分別為92.65%、88.24%以及81.86%，混合三者孕婦心電訊號訓練與測試得到平均偵測率為91.83%，相較於其他文獻使用的K近鄰演算法，卷積神經網路在偵測上獲得較良好的效果。為了達到穿戴式裝置低功耗的目標，進行了硬體的設計與實現。在硬體設計上，為了得到較佳的產出率與硬體複雜度之折衷方案，我們採用了管線化之設計，為了降低硬體面積大小，在兩層卷積層(convolutional layer)與全連接層中(fully connected layer)共使用235個乘法器，並透過排程與硬體架構設計來降低對外部記憶體資料的提取以及提升硬體使用效率來達到省時省面積之目的。

In this thesis, we present a recognition and monitoring system for detecting fetal Electrocardiogram (ECG) from pregnant abdominal ECG recordings. The deep learning algorithm, 2-dimensional (2D) convolutional neural network, is used to realize the classification. With the non-invasive electrodes, our system can detect long-term fetal ECG signals accurately. Thus, the growth of health condition of the baby can be monitored with the wearable device. The abdominal ECG waveform is partitioned into 250ms segments with an extra overlapping 20ms interval at the beginning and at the end of each segment to reserve the signal integrity. In order to reduce the noise effect of the ECG signals in the time domain, short-time Fourier transform (STFT) is adopt which transforms one-dimensional ECG signals into two-dimensional time-frequency representation. Consequently, the 2D convolutional neural network can extract the feature on the time-frequency plane to enhance the classification. Because 4 leads are used to measure and record maternal abdominal ECG, fusion is required in the last phase. The probabilities of respective classes judged from different electrodes are then combined to generate the final results. The individual detection rates of pregnant r01, r08 and r10 are 92.65% 88.24% and 81.86%, respectively. The average detection rate achieves 91.83% if three pregnant ECG signals are mixed for training and inference. Compared with the conventional K-nearest neighbor algorithm, a higher detection accuracy is achieved. For hardware implementation, pipeline architecture is used to tradeoff throughput and complexity. To save silicon area, 235 multipliers are adopted for the two convolution layers and one fully-connected layer. The architecture is designed with proper scheduling to reduce the global memory access and to increase hardware utilization.

[1] V. Ionescu and M. Hnatiuc, "Fetal heart rate detection and monitoring from noninvasive abdominal ECG recordings," 2015 E-Health and Bioengineering Conference (EHB), Iasi, 2015, pp. 1-4.
[2] J. A. Delgado, M. Altuve and M. N. Homsi, "Haar wavelet transform and principal component analysis for fetal QRS classification from abdominal maternal ECG recordings," 2015 20th Symposium on Signal Processing, Images and Computer Vision (STSIVA), Bogota, 2015, pp. 1-6.
[3] J. Cardenas-Lattus and H. Kaschel, "Fetal ECG multi-level analysis using daubechies wavelet transform for non-invasive maternal abdominal ECG recordings," 2017 CHILEAN Conference on Electrical, Electronics Engineering, Information and Communication Technologies (CHILECON), Pucon, 2017, pp. 1-6.
[4] J. W. Jhuang and H. P. Ma, “A Patch-sized Wearable ECG/Respiration Recording Platform with DSP Capability,” 17th International Conference on E-health Networking, Application & Services (HealthCom), 2015.
[5] F. Andreotti, F. Gräßer, H. Malberg and S. Zaunseder, "Non-invasive Fetal ECG Signal Quality Assessment for Multichannel Heart Rate Estimation," in IEEE Transactions on Biomedical Engineering, vol. 64, no. 12, pp. 2793-2802, Dec. 2017.
[6] L. Han, X. j. Pu and X. j. Chen, "Method of fetal electrocardiogram extraction based on ν-support vector regression," in IET Signal Processing, vol. 9, no. 5, pp. 430-439, 7 2015.
[7] P. Kumar, S. K. Sharma and S. Prasad, "Detection of fetal electrocardiogram through ofdm, neuro-fuzzy logic and wavelets systems for telemetry," 2016 10th International Conference on Intelligent Systems and Control (ISCO), Coimbatore, 2016, pp. 1-4.
[8] Cardiovascular Physiology Concepts http://www.cvphysiology.com/
[9] I. Silva, J. Behar, R. Sameni, T. Zhu, J. Oster, G. D. Clifford, and G. B. Moody. Noninvasive fetal ecg: the physionet/computing in cardiology challenge 2013. In Computing in Cardiology Conference (CinC), 2013, pages 149–152. IEEE, 2013.
[10] Jezewski J, Matonia A, Kupka T, Roj D, Czabanski R. >Determination of the fetal heart rate from abdominal signals: evaluation of beat-to-beat accuracy in relation to the direct fetal electrocardiogram.Biomedical Engineering/Biomedizinische Technik 2012 Jul;57(5):383-394. [11] V. Starc, "Non-invasive fetal multilead RR interval determination from maternal abdominal recordings: The Physionet/CinC Challenge 2013," Computing in Cardiology 2013, Zaragoza, 2013, pp. 317-320.
[12] Institute of medical technology and equipment http://www.itam.zabrze.pl/index.php/
[13] S. Kiranyaz, T. Ince and M. Gabbouj, "Real-Time Patient-Specific ECG Classification by 1-D Convolutional Neural Networks," in IEEE Transactions on Biomedical Engineering, vol. 63, no. 3, pp. 664-675, March 2016.
[14] S. Y. Chun, J. H. Kang, H. Kim, C. Lee, I. Oakley and S. P. Kim, "ECG based user authentication for wearable devices using short time Fourier transform," 2016 39th International Conference on Telecommunications and Signal Processing (TSP), Vienna, 2016, pp. 656-659.
[15] I. Odinaka et al., "ECG biometrics: A robust short-time frequency analysis," 2010 IEEE International Workshop on Information Forensics and Security, Seattle, WA, 2010, pp. 1-6.
[16] K. Fukushima, "Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position," Biological Cybernetics, vol. 36, pp. 193-202, 1980.
[17] Sean shensheng Xu, Man-Wai Mak and Chi-Chung Cheung, "Deep neural networks versus support vector machines for ECG arrhythmia classification," 2017 IEEE International Conference on Multimedia & Expo Workshops (ICMEW), Hong Kong, 2017, pp. 127-132.
[18] V. César Cavalcanti Roza, A. M. de Almeida and O. Adrian Postolache, "Design of an artificial neural network and feature extraction to identify arrhythmias from ECG," 2017 IEEE International Symposium on Medical Measurements and Applications (MeMeA), Rochester, MN, 2017, pp. 391-396.
[19] M. Cheng, W. J. Sori, F. Jiang, A. Khan and S. Liu, "Recurrent Neural Network Based Classification of ECG Signal Features for Obstruction of Sleep Apnea Detection," 2017 IEEE International Conference on Computational Science and Engineering (CSE) and IEEE International Conference on Embedded and Ubiquitous Computing (EUC), Guangzhou, 2017, pp. 199-202.
[20] Y. Lecun, L. Bottou, Y. Bengio and P. Haffner, "Gradient-based learning applied to document recognition," in Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, Nov 1998.
[21] M. Bacis, G. Natale, E. Del Sozzo and M. D. Santambrogio, "A pipelined and scalable dataflow implementation of convolutional neural networks on FPGA," 2017 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), Lake Buena Vista, FL, 2017, pp. 90-97.
[22] M. Zhu, Q. Kuang, C. Yang and J. Lin, "Optimization of convolutional neural network hardware structure based on FPGA," 2018 13th IEEE Conference on Industrial Electronics and Applications (ICIEA), Wuhan, China, 2018, pp. 1797-1802.