簡易檢索 / 詳目顯示

研究生: 余自祥
Zhi-Siong Yii
論文名稱: 基於卷積神經網路與長短期記憶之瞳孔光反射量測
Pupillary Light Reflex Measurement Based on Convolution Neural Network and Long Short-Term Memory
指導教授: 陳怡永
Yi-Yung Chen
黃忠偉
Allen Jong-Woei Whang
口試委員: 趙一平
Yi-Ping Chao
歐立成
Li-Chen Ou
孫沛立
Pei-Li Sun
陳怡永
Yi-Yung Chen
黃忠偉
Allen Jong-Woei Whang
學位類別: 碩士
Master
系所名稱: 應用科技學院 - 色彩與照明科技研究所
Graduate Institute of Color and Illumination Technology
論文出版年: 2022
畢業學年度: 110
語文別: 中文
論文頁數: 93
中文關鍵詞: 人工智慧雙向長短期記憶卷積神經網路深度學習長短期記憶便攜式瞳孔計瞳孔光反射
外文關鍵詞: artificial intelligence, bidirectional long short-term memory, convolution neural network, deep learning, long short-term memory, portable pupillometer, pupillary light reflex
相關次數: 點閱:381下載:0
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報
  • 瞳孔大小會隨著進入眼睛的光量而變化,這個過程稱為瞳孔光反射。醫護人員通過觀測瞳孔光反射來評估患者的生理狀態,病情和找尋病因。在臨床上,醫護人員通過觀察因筆燈的光刺激而促使的瞳孔收縮的幅度與速度,判斷出瞳孔光反射的等級。由於瞳孔光反射的分級缺乏定量的定義與標準的臨床協議,再加上醫護人員經驗水平的不同或其他因素的影響,醫護人員依靠視覺觀測的主觀評估存在著評估者間與評估者內的評估一致性不足的問題。這也導致醫護人員無法正確追蹤並瞭解患者的狀況。

    本研究提出了一種基於卷積神經網路與長短期記憶或雙向長短期記憶建構的深度學習演算法,實現客觀的自動化瞳孔光反射量測。本研究與臺灣林口長庚紀念醫院的兩位眼科醫師合作,通過臨床試驗獲取資料集。本研究使用團隊開發的便攜式瞳孔計獲得記錄了受試者的瞳孔光反射影片,並將影片作為資料集。瞳孔光反射的分級標籤由合作的兩位眼科醫師通過觀察影片後提供,作為資料集的基本事實。

    本研究使用人工智慧技術的深度學習模型從資料集中學習兩位眼科醫師潛在的分級標準與規則,從而實現更準確與可重複的瞳孔光反射量測,輔助醫師進行診斷,減輕醫護人員的負擔並提高醫療品質。

    本研究使用通過臨床試驗得到的124筆資料,訓練出三種分級方式(精細分級,粗略分級和二元分級)的模型。最後,依據模型的測試結果得知,粗略分級與二元分級的模型更能勝任這份工作。在未來,可將模型移植到便攜式瞳孔計上,實現瞳孔光反射量測功能,方便醫護人員迅速地爲受試者進行瞳孔光反射的檢查。


    Pupil size changes with the amount of light entering the eye, this process is called Pupillary Light Reflex. Medical staff evaluate the patient's physiological state, and condition and find the cause by observing the pupillary light reflex. Clinically, medical staff judge the grade of pupillary light reflex by observing the amplitude and speed of pupil constriction induced by the light stimulation of the penlight. Due to the lack of quantitative definitions and standard clinical protocols for the pupillary light reflex measurement, coupled with differences in the experience of medical staff or other factors, the medical staff's subjective assessment relying on visual observations has the problems with the interrater and intrarater reliability. This also results in the inability of the medical staff to properly track and understand the patient's condition.

    This study proposes a deep learning model based on Convolutional Neural Network and Long Short-Term Memory or Bidirectional Long-Short-Term Memory to achieve an objective and automated pupillary light reflex measurement. This study cooperates with two ophthalmologists at Chang Gung Memorial Hospital in Linkou, Taiwan to obtain the clinical trial dataset. For this study, a portable pupillometer developed by our team is used to obtain the videos recording the pupillary light reflexes. The videos are served as the dataset. The graded labels of pupillary light reflexes are provided by two collaborating ophthalmologists after watching the videos. The graded labels are served as the ground truth for the dataset.

    This study uses the deep learning model of artificial intelligence technology to learn the potential grading standards and rules of two ophthalmologists from the dataset. It can achieve a more accurate and repeatable pupillary light reflex measurement to assist physicians in the diagnosis, reduce the burden on the medical staff, and improve the quality of medical care.

    This study uses 124 data that is obtained through the clinical trials to train the models with three grading methods (fine grading, coarse grading, and binary grading). Finally, according to the test results of the models, it is known that the coarse grading and binary grading models are more suitable for the job. In the future, the model can be transplanted to the portable pupillometer to realize the function of pupillary light reflex measurement, which is convenient for the medical staff to perform a pupillary light reflex examination for subjects quickly.

    摘要 I Abstract II 誌謝 IV 目錄 V 圖目錄 VIII 表目錄 X 第一章 緒論 1 1.1 研究背景 1 1.2 研究動機 3 1.3 研究目的 6 1.4 論文架構 7 第二章 文獻探討 9 2.1 眼睛 9 2.2 眼睛構造 10 2.3 瞳孔 12 2.4 瞳孔收縮與擴張 12 2.5 瞳孔光反射 14 2.6 加權kappa統計 15 2.7 人工智慧 15 2.7.1 機器學習 16 2.7.2 深度學習 17 2.8 機器學習與深度學習之差異 17 2.9 深度學習模型之擬合 19 2.10 卷積神經網路 20 2.10.1 卷積層 21 2.10.2 池化層 22 2.10.3 扁平層 24 2.10.4 全連接層 24 2.10.5 激勵函數 25 2.10.6 輸出層 25 2.10.7 損失函數 26 2.11 遷移學習 27 2.12 Grad-cam++ 28 2.13 長短期記憶與雙向長短期記憶 28 第三章 研究方法 32 3.1 深度學習模型之架構 32 3.2 模型之訓練與學習 37 3.3 臨床試驗 39 3.4 資料擴增 43 3.5 瞳孔光反射資料集 45 3.6 模型超參數調整 49 3.7 研究設備 51 第四章 研究結果與討論 52 4.1 模型的測試 52 4.2 預測錯誤的原因分析 56 4.3 評估者間與評估者內可靠性 58 4.4 模型運算速度 59 第五章 結論與未來展望 60 5.1 結論 60 5.2 未來展望 61 參考文獻 63 附錄一、臨床試驗與研究同意證明書 67 附錄二、受試者同意書 74

    [1] Lowenstein, O., & Loewenfeld, I. E. (1950). Role of sympathetic and parasympathetic systems in reflex dilatation of the pupil: Pupillographic studies. Archives of Neurology & Psychiatry, 64(3), 313-340.
    [2] Ettinger, E. R., Wyatt, H. J., & London, R. (1991). Anisocoria. Variation and clinical observation with different conditions of illumination and accommodation. Investigative ophthalmology & visual science, 32(3), 501-509.
    [3] Carrick, F. R., Azzolino, S. F., Hunfalvay, M., Pagnacco, G., Oggero, E., D’Arcy, R. C., ... & Sugaya, K. (2021). The pupillary light reflex as a biomarker of concussion. Life, 11(10), 1104.
    [4] Belliveau, A. P., Somani, A. N., & Dossani, R. H. (2019). Pupillary light reflex.
    [5] Zhang, F., Kurokawa, K., Lassoued, A., Crowell, J. A., & Miller, D. T. (2019). Cone photoreceptor classification in the living human eye from photostimulation-induced phase dynamics. Proceedings of the National Academy of Sciences, 116(16), 7951-7956.
    [6] Curcio, C. A., & Allen, K. A. (1990). Topography of ganglion cells in human retina. Journal of comparative Neurology, 300(1), 5-25.
    [7] Pfeiffer, R. L., Anderson, J. R., Dahal, J., Garcia, J. C., Yang, J. H., Sigulinsky, C. L., ... & Jones, B. W. (2020). A pathoconnectome of early neurodegeneration: Network changes in retinal degeneration. Experimental eye research, 199, 108196.
    [8] Pickard, G. E., & Sollars, P. J. (2011). Intrinsically photosensitive retinal ganglion cells. Reviews of physiology, biochemistry and pharmacology, 59-90.
    [9] Ebitz, R. B., & Moore, T. (2017). Selective modulation of the pupil light reflex by microstimulation of prefrontal cortex. Journal of Neuroscience, 37(19), 5008-5018.
    [10] Pinheiro, H. M., & da Costa, R. M. (2021). Pupillary light reflex as a diagnostic aid from computational viewpoint: A systematic literature review. Journal of Biomedical Informatics, 117, 103757.
    [11] Mathew, A., Amudha, P., & Sivakumari, S. (2020, February). Deep learning techniques: an overview. In International conference on advanced machine learning technologies and applications (pp. 599-608). Springer, Singapore.
    [12] Chen, J. W., Gombart, Z. J., Rogers, S., Gardiner, S. K., Cecil, S., & Bullock, R. M. (2011). Pupillary reactivity as an early indicator of increased intracranial pressure: the introduction of the Neurological Pupil index. Surgical neurology international, 2.
    [13] Park, J. G., Moon, C. T., Park, D. S., & Song, S. W. (2015). Clinical utility of an automated pupillometer in patients with acute brain lesion. Journal of Korean Neurosurgical Society, 58(4), 363-367.
    [14] Kramer, C. L., Rabinstein, A. A., Wijdicks, E. F., & Hocker, S. E. (2014). Neurologist versus machine: is the pupillometer better than the naked eye in detecting pupillary reactivity. Neurocritical care, 21(2), 309-311.
    [15] Meeker, M., Du, R., Bacchetti, P., & Privitera, C. M. (2005). Pupil examination: validity and clinical utility of an automated pupillometer. Journal of Neuroscience Nursing, 37(1), 34.
    [16] Wilson, S. F., Amling, J. K., Floyd, S. D., & McNair, N. D. (1988). Determining interrater reliability of nurses' assessments of pupillary size and reaction. The Journal of neuroscience nursing: journal of the American Association of Neuroscience Nurses, 20(3), 189-192.
    [17] Van den Berge, J. H., Schouten, H. J., Boomstra, S., van Drunen Littel, S., & Braakman, R. (1979). Interobserver agreement in assessment of ocular signs in coma. Journal of Neurology, Neurosurgery & Psychiatry, 42(12), 1163-1168.
    [18] Koo, T. K., & Li, M. Y. (2016). A guideline of selecting and reporting intraclass correlation coefficients for reliability research. Journal of chiropractic medicine, 15(2), 155-163.
    [19] Willoughby, C. E., Ponzin, D., Ferrari, S., Lobo, A., Landau, K., & Omidi, Y. (2010). Anatomy and physiology of the human eye: effects of mucopolysaccharidoses disease on structure and function–a review. Clinical & Experimental Ophthalmology, 38, 2-11.
    [20] Spector, R. H. (1990). The pupils. Clinical Methods: The History, Physical, and Laboratory Examinations. 3rd edition.
    [21] Packiasabapathy, S., Rangasamy, V., & Sadhasivam, S. (2021). Pupillometry in perioperative medicine: a narrative review. Canadian Journal of Anesthesia/Journal canadien d'anesthésie, 68(4), 566-578.
    [22] Kierkegaard, M., & Tollbäck, A. (2005). Inter-and intra-rater reliability of the B. Lindmark Motor Assessment. Advances in Physiotherapy, 7(1), 2-6.
    [23] Yang, Z., & Zhou, M. (2015). Weighted kappa statistic for clustered matched-pair ordinal data. Computational Statistics & Data Analysis, 82, 1-18.
    [24] McHugh, M. L. (2012). Interrater reliability: the kappa statistic. Biochemia medica, 22(3), 276-282.
    [25] Bini, S. A. (2018). Artificial intelligence, machine learning, deep learning, and cognitive computing: what do these terms mean and how will they impact health care?. The Journal of arthroplasty, 33(8), 2358-2361.
    [26] Nichols, J. A., Herbert Chan, H. W., & Baker, M. A. (2019). Machine learning: applications of artificial intelligence to imaging and diagnosis. Biophysical reviews, 11(1), 111-118.
    [27] Hussain, M., Bird, J. J., & Faria, D. R. (2018, September). A study on cnn transfer learning for image classification. In UK Workshop on computational Intelligence (pp. 191-202). Springer, Cham.
    [28] Manaswi, N. K. (2018). CNN in Keras. In Deep Learning With Applications Using Python (pp. 105-114). Apress, Berkeley, CA.
    [29] Agarap, A. F. (2017). An architecture combining convolutional neural network (CNN) and support vector machine (SVM) for image classification. arXiv preprint arXiv:1712.03541.
    [30] Abdel-Hamid, O., Mohamed, A. R., Jiang, H., Deng, L., Penn, G., & Yu, D. (2014). Convolutional neural networks for speech recognition. IEEE/ACM Transactions on audio, speech, and language processing, 22(10), 1533-1545.
    [31] Dewa, C. K. (2018). Suitable CNN weight initialization and activation function for Javanese vowels classification. Procedia computer science, 144, 124-132.
    [32] Deng, J., Dong, W., Socher, R., Li, L. J., Li, K., & Fei-Fei, L. (2009, June). Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition (pp. 248-255). Ieee.
    [33] Chattopadhay, A., Sarkar, A., Howlader, P., & Balasubramanian, V. N. (2018, March). Grad-cam++: Generalized gradient-based visual explanations for deep convolutional networks. In 2018 IEEE winter conference on applications of computer vision (WACV) (pp. 839-847). IEEE.
    [34] Koutnik, J., Greff, K., Gomez, F., & Schmidhuber, J. (2014, June). A clockwork rnn. In International Conference on Machine Learning (pp. 1863-1871). PMLR.
    [35] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
    [36] Liu, Y., Wang, Y., Yang, X., & Zhang, L. (2017, October). Short-term travel time prediction by deep learning: A comparison of different LSTM-DNN models. In 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC) (pp. 1-8). IEEE.
    [37] Huang, Z., Xu, W., & Yu, K. (2015). Bidirectional LSTM-CRF models for sequence tagging. arXiv preprint arXiv:1508.01991.
    [38] Crisóstomo de Castro Filho, H., Abílio de Carvalho Júnior, O., Ferreira de Carvalho, O. L., Pozzobon de Bem, P., dos Santos de Moura, R., Olino de Albuquerque, A., ... & Trancoso Gomes, R. A. (2020). Rice crop detection using LSTM, Bi-LSTM, and machine learning models from sentinel-1 time series. Remote Sensing, 12(16), 2655.
    [39] Ullah, A., Ahmad, J., Muhammad, K., Sajjad, M., & Baik, S. W. (2017). Action recognition in video sequences using deep bi-directional LSTM with CNN features. IEEE access, 6, 1155-1166.
    [40] Graves, A., & Schmidhuber, J. (2005). Framewise phoneme classification with bidirectional LSTM and other neural network architectures. Neural networks, 18(5-6), 602-610.
    [41] Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L. C. (2018). Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 4510-4520).
    [42] Aslan, M. F., Unlersen, M. F., Sabanci, K., & Durdu, A. (2021). CNN-based transfer learning–BiLSTM network: A novel approach for COVID-19 infection detection. Applied Soft Computing, 98, 106912.
    [43] Moharm, K., Eltahan, M., & Elsaadany, E. (2020, November). Wind Speed Forecast using LSTM and Bi-LSTM Algorithms over Gabal El-Zayt Wind Farm. In 2020 International Conference on Smart Grids and Energy Systems (SGES) (pp. 922-927). IEEE.

    無法下載圖示 全文公開日期 2025/08/27 (校內網路)
    全文公開日期 2025/08/27 (校外網路)
    全文公開日期 2025/08/27 (國家圖書館:臺灣博碩士論文系統)
    QR CODE