應用卷積神經網路與生成對抗網路於急性白血病分類預測之研究

Acute Leukemia Prediction and Classification using Convolutional Neural Networks and Generative Adversarial Networks

魏吉宏(Chi-Hung Wei)

指導教授 ： 連俊瑋

共同指導教授 ： 陳牧言(Mu-Yen Chen)

國立臺中科技大學/資訊與流通學院/資訊管理系碩士班/碩士(2022年)

摘要

白血病是影響血液細胞的癌症一種，如果沒有及早的發現與治療，會造成致命的結果，而急性白血病的檢驗方法中，流式細胞分析術是重要的工具之一，但其龐大的數據量，和依賴於專業人員使用人工圈選方法，往往費時又費工。本研究使用深度學習的卷積神經網路方法(Convolutional Neural Network, CNN)，對流式細胞數據參數組合的二維圖形來做分類預測，辨識急性白血病類型為非急性白血病、急性骨髓白血病和急性淋巴白血病三種分類。目的以原始數據、數據正規化後、正規化後移除離群值及經生成對抗網路(Generative Adversarial Network, GAN)增生資料，四種方式所產生之不同參數組合的二維散佈圖，經訓練後之分類預測準確率，何種方式及何種參數組合效果最佳。
通過人體試驗倫理委員會申請，取得近五年500筆的流式細胞初級分析的八色急性白血病篩選單管(Acute Leukemia Orientation Tube, ALOT)資料，每一資料中包含了25萬個細胞數據，數據參數包含了細胞大小和抗原反應等12個參數。在資料前處理，分成了四種資料集，再各自生成12種不同參數組合的二維散佈圖，使用CNN方法來訓練和分類預測。本研究實驗後得到了準確率73%到86%的成果，在四種方式產生之不同參數組合二維散佈圖中，均以CD3和CD7細胞群抗原參數之組合分類預測準確率最佳，且以GAN資料增生後之圖片資料集分類辨識準確率達到86%為最高。運用深度學習方法在檢驗數據日趨增加的現在，執行大量且客觀的重複性分析，減少因人為操作所帶來的缺點，快速得到檢驗結果，盡早確定疾病類型並進行治療。

關鍵字

卷積神經網路 ； 正規化 ； 生成對抗網路 ； 急性白血病 ； 流式細胞分析

並列摘要

Leukemia is a type of cancer that affects blood cells and can be fatal if not detected and treated early. In the detection method of acute leukemia, flow cytometry is one of the important tools. But its huge data and reliance on professionals to use manual gatting methods are often time-consuming and labor-intensive. This study uses the deep learning Convolutional Neural Network (CNN) method to make classification and prediction on the two-dimensional graph of the parameter combination of flow cytometry data. Identify the classes of acute leukemia into three categories: non-acute leukemia, acute myeloid leukemia and acute lymphoblastic leukemia. The purpose is to use the original data, normalized data, removed outliers after normalization, and augmented data by Generative Adversarial Network (GAN), two-dimensional scatter plots of different parameter combinations generated by four methods. The classification prediction accuracy after training, which method and which parameter combination is the best.
Through the application of Institutional Review Board (IRB), 500 pieces of 8-color Acute Leukemia Orientation Tube (ALOT) data for primary analysis of flow cytometry in the past five years were obtained. Each data contains 250,000 cell data, and the data parameters include 12 parameters such as cell size and antigen response. In the pre-processing of the data, it is divided into four data sets, and then each generates a two-dimensional scatter plot of 12 different parameter combinations, and uses the CNN method for training and classification prediction. After the experiment in this study, the accuracy rate of 73% to 86% was obtained. In the two-dimensional scattergrams of different parameter combinations generated by the four methods, the combination of CD3 and CD7 cell population parameters has the best classification and prediction accuracy, and the classification and identification accuracy of the image data set after the GAN data is proliferated reaches 86%. to the highest. With the increasing number of inspection data, deep learning methods are used to perform a large number of objective and repetitive analyses, reduce the shortcomings caused by manual operations, obtain inspection results quickly, and determine the type of disease and treat it as soon as possible.

並列關鍵字

Convolutional Neural Networks ； Generative Adversarial Networks ； Z-score normalizaiton ； Acute Leukemia ； Flow Cytometry

參考文獻

1. Abhishek, A., Jha, R. K., Sinha, R., Jha, K. (2022). Automated classification of acute leukemia on a heterogeneous dataset using machine learning and deep learning techniques. Biomedical Signal Processing and Control, 72, 103341.

2. Aggarwal, A., Mittal, M., Battineni, G. (2021). Generative adversarial network: An overview of theory and applications. International Journal of Information Management Data Insights, 1(1), 100004.

3. Aghaeepour, N., Nikolic, R., Hoos, H. H., Brinkman, R. R. (2011). Rapid cell population identification in flow cytometry data. Cytometry Part A, 79(1), 6-13. Retrieved from https://onlinelibrary.wiley.com/doi/full/10.1002/cyto.a.21007

4. Ahmed, N., Yigit, A., Isik, Z., Alpkocak, A. (2019). Identification of leukemia subtypes from microscopic images using convolutional neural network. Diagnostics, 9(3), 104.

5. Alam, A., Anwar, S. (2021). Detecting Acute Lymphoblastic Leukemia Through Microscopic Blood Images Using CNN. Trends in Wireless Communication and Information Security, 207-214.