病患在麻醉誘導後發生低血壓是常見的問題，若在手術期間發生低血壓，可能在術後引發併發症。而目前有研究利用深度學習模型預測病患未來是否發生低血壓以協助醫療人員採取適當的醫療方針進而降低病患的風險。由於動脈血壓波形是一個可呈現出血管對血液的調節、心臟收縮能力等心血管系統狀態的重要參數，因此我們想利用深度學習的方式，透過病患在麻醉誘導前的動脈血壓波形，預測患者未來發生低血壓的機率。本研究使用了VitalDB資料庫中的23名患者在麻醉誘導前後的動脈血壓波形和相關資料。與過去的研究不同，我們選擇使用離散小波轉換，將動脈血壓波形轉換為包含不同頻帶中能量變化的時頻圖，以避免繁複的計算和耗時的測量。接著利用卷積神經網路對這些時頻圖進行自動提取特徵和預測。考慮到案例的數量有限，我們額外劃分出測試集以評估該模型的泛化能力，並使用AUROC和AUPRC指標進行性能評估。總體而言，本研究所提出的方法旨在提取動脈血壓波形中不同頻帶之間的關係，並將其作為病患「生理狀態」的特徵進行分類以預測是否在麻醉誘導後發生低血壓。最後，得到了AUROC與AUPRC皆高於0.9的表現。

Post-induction hypotension is a common problem and may lead to postoperative complications if it happens during surgery. Currently, there are studies utilizing deep learning models to predict hypotension in the future, which helps clinicians adopt appropriate medical strategies to reduce patient risks. Arterial blood pressure waveform is an important parameter that reflects the regulation of blood vessels, cardiac contractility, and other cardiovascular system conditions. Therefore, we aim to use deep learning model to predict post-induction hypotension based on their arterial blood pressure waveform before anesthesia induction. In this study, we used arterial blood pressure waveforms and related data from 23 patients in the VitalDB database. Compare to previous research, we chose to use discrete wavelet transform to convert arterial blood pressure waveforms into time-frequency spectrograms containing energy variations in different frequency bands, to avoid complex calculations and time-consuming measurements. Then, we employed a convolutional neural network to automatically extract features from these spectrograms and make predictions. Considering the limited number of cases, we separated a testing set to evaluate the generalization of the model, and used AUROC and AUPRC for performance evaluation. Overall, the method proposed in this study aims to extract the relationships between different frequency bands in arterial blood pressure waveforms and use them as features to classify patients' physiological states for predicting the occurrence of post-induction hypotension. Finally, we achieved performance with AUROC and AUPRC both exceeding 0.9.

[1] K. Maheshwari et al., "The association of hypotension during non-cardiac surgery, before and after skin incision, with postoperative acute kidney injury: a retrospective cohort analysis," (in English), Anaesthesia, Article vol. 73, no. 10, pp. 1223-1228, Oct 2018, doi: 10.1111/anae.14416.
[2] T. G. Monk et al., "Association between Intraoperative Hypotension and Hypertension and 30-day Postoperative Mortality in Noncardiac Surgery," (in English), Anesthesiology, Article vol. 123, no. 2, pp. 307-319, Aug 2015, doi: 10.1097/aln.0000000000000756.
[3] C. S. Lin, J. S. Chiu, M. H. Hsieh, M. S. Mok, Y. C. Li, and H. W. Chiu, "Predicting hypotensive episodes during spinal anesthesia with the application of artificial neural networks," Computer Methods and Programs in Biomedicine, vol. 92, no. 2, pp. 193-197, Nov 2008, doi: 10.1016/j.cmpb.2008.06.013.
[4] J. D. Bristow, C. Prys-Roberts, A. Fisher, Thomas G. Pickering, and P. Sleight, "Effects of Anesthesia on Baroreflex Control of Heart Rate in Man," Anesthesiology, vol. 31, no. 5, pp. 422-428, 1969, doi: 10.1097/00000542-196911000-00009.
[5] J. Marty and J. G. Reves, "CARDIOVASCULAR CONTROL MECHANISMS DURING ANESTHESIA," Anesthesia and Analgesia, vol. 69, no. 3, pp. 273-275, Sep 1989. [Online]. Available: <Go to ISI>://WOS:A1989AN25800001.
[6] M. R. Pinsky and A. Dubrawski, "Gleaning knowledge from data in the intensive care unit," Am J Respir Crit Care Med, vol. 190, no. 6, pp. 606-10, Sep 15 2014, doi: 10.1164/rccm.201404-0716CP.
[7] S. Laurent and P. Boutouyrie, "Recent Advances in Arterial Stiffness and Wave Reflection in Human Hypertension," Hypertension, vol. 49, no. 6, pp. 1202-1206, 2007, doi: doi:10.1161/HYPERTENSIONAHA.106.076166.
[8] S. Wright et al., "Colour Doppler ultrasound of the ocular circulation in patients with systemic lupus erythematosus identifies altered microcirculatory haemodynamics," Lupus, vol. 18, no. 11, pp. 950-957, 2009, doi: 10.1177/0961203309104865.
[9] L. Keselbrener and S. Akselrod, "Time-frequency analysis of transient signals - application to cardiovascular control," Physica A, vol. 249, no. 1-4, pp. 482-490, Feb 1998, doi: 10.1016/s0378-4371(97)00507-4.
[10] M. Malik et al., "Heart rate variability: Standards of measurement, physiological interpretation, and clinical use," Eur Heart J, vol. 17, no. 3, pp. 354-381, 1996, doi: 10.1093/oxfordjournals.eurheartj.a014868.
[11] R. Vetter et al., "Subband modeling of the human cardiovascular system: New insights into cardiovascular regulation," Annals of Biomedical Engineering, vol. 26, no. 2, pp. 293-307, Mar-Apr 1998, doi: 10.1114/1.57.
[12] C. E. Agnew, A. J. McCann, C. J. Lockhart, P. K. Hamilton, G. E. McVeigh, and R. C. McGivern, "Comparison of RootMUSIC and Discrete Wavelet Transform Analysis of Doppler Ultrasound Blood Flow Waveforms to Detect Microvascular Abnormalities in Type I Diabetes," Ieee Transactions on Biomedical Engineering, vol. 58, no. 4, pp. 861-867, Apr 2011, doi: 10.1109/tbme.2010.2097263.
[13] A. Pachauri and M. Bhuyan, "The Wavelet Transform Based Arterial Blood Pressure Waveform Delineator," 2012.
[14] M. De Melis et al., "Blood pressure waveform analysis by means of wavelet transform," Medical & Biological Engineering & Computing, vol. 47, no. 2, pp. 165-173, Feb 2009, doi: 10.1007/s11517-008-0397-9.
[15] S. Mahmoodabadi, A. Ahmadian, M. Abolhasani, M. Eslami, and J. Bidgoli, "ECG Feature Extraction Based on Multiresolution Wavelet Transform," (in eng), Conf Proc IEEE Eng Med Biol Soc, vol. 2005, pp. 3902-5, 2005, doi: 10.1109/iembs.2005.1615314.
[16] F. Hatib et al., "Machine-learning Algorithm to Predict Hypotension Based on High-fidelity Arterial Pressure Waveform Analysis," Anesthesiology, vol. 129, no. 4, pp. 663-674, Oct 2018, doi: 10.1097/aln.0000000000002300.
[17] Y. Y. Jo et al., "Predicting intraoperative hypotension using deep learning with waveforms of arterial blood pressure, electroencephalogram, and electrocardiogram: Retrospective study," Plos One, vol. 17, no. 8, Aug 2022, doi: 10.1371/journal.pone.0272055.
[18] J. Lee et al., "Comparative Analysis on Machine Learning and Deep Learning to Predict Post-Induction Hypotension," (in English), Sensors, Article vol. 20, no. 16, p. 20, Aug 2020, Art no. 4575, doi: 10.3390/s20164575.
[19] H. C. Lee, Y. Park, S. B. Yoon, S. M. Yang, D. Park, and C. W. Jung, "VitalDB, a high-fidelity multi-parameter vital signs database in surgical patients," (in eng), Sci Data, vol. 9, no. 1, p. 279, Jun 8 2022, doi: 10.1038/s41597-022-01411-5.
[20] S. Ioffe and C. Szegedy, "Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift," in International Conference on Machine Learning, 2015.