應用深度暨機器學習方法對肺部X光進行病灶分類之研究

Classified Lesions of Chest X-ray Images Using Deep and Machine Learning Methods

宋建軒(Jian-Hsuan Sung)

指導教授 ： 陳泰賓

義守大學/智慧科技學院/資訊工程學系/碩士(2021年)

摘要

動機與目的： 胸腔X光檢查在臨床上是重要及最普遍的檢查之一。胸部X光影像的診斷與判讀，取決於操作者暨醫師診斷之技術與經驗。近幾年人工智慧輔助影像判讀之研究興盛，除判讀病灶之外，影像品質判讀與分類為降低誤判率之關鍵。利用人工智慧深度學習輔助影像品質判讀，提升診斷效率。期望依據此模型可用於輔助深度學習影像判讀分類影像病灶之工具，降低誤判機率與提供判讀者於影像判讀時之輔助工具。
材料與方法：本研究使用之胸部X光影像取自美國國家衛生院(National Institutes of Health, NIH)公開資料庫中十萬張胸部X光影像，並從中篩選出五類常見胸部X光疾病；分別是積水(Effusion)、肺炎(Pneumonia)浸潤(Infiltrate)、腫塊(Mass)及結節(Nodule)，各選50張總共250張。利用Matlab程式建立深度學習網路模型(Convolution Neural Network, CNN)進行四種不同的影像分類，分別為瀰漫性與局部性、積水與肺炎、腫塊(Mass)與結節及積水與浸潤，以三種監督式模型訓練神經網路，分別為ALexNet、VGG19及ResNet50，再結合三種機器學習，分別為支持向量機(Support Vector Machine, SVM)、線性回歸(Linear Regression)及單純貝氏分類器(Naive Bayes)，驗證模型效能評估以訓練組的準確性(Accuracy, ACC)、特異性(Specificity, SPE)、靈敏度(Sensitivity, SEN)、陽性預測值(Positive Predicted Value, PPV)、陰性預測值(Negative Predicted Value, NPV)及 Kappa 一致性統計量。
結果：ResNet-50搭配NaiveBayes 分類模型在測試集之分類效能在所有組合中具有最好的表現；其準確度、靈敏度、特異性、Kappa 分別為 0.93、0.95、0.90、0.857。
結論：ALexNet、ResNet-50 與 VGG19 均可搭配機器學習建立影像分類模型。根據本研究結果顯示，利用這三種 CNN 模型搭配機器學習對胸部X光影像進行分類具有可行性。

關鍵字

人工智慧 ； 深度學習演算法 ； 卷積神經網路 ； 影像分類 ； 胸部X光影像

並列摘要

The diagnosis of chest X-ray images depends on the operator’s skill and the doctor’s experience. In recent years, researches on the use of artificial intelligence technology for image interpretation have prospered in researches such as lesion identification. The diagnostic accuracy by using artificial intelligence was highly based on the quality of chest X-ray images. Therefore, this study uses artificial intelligence deep learning to distinguish and classify image lesions.
Using 100,000 chest X-ray images from the National Institutes of Health (NIH) database and screen out four types of common chest X-ray images are for experiments, including hydrops, pneumonia, infiltration, and mass. There are four types of chest X-ray images of chest and nodules. Convolutional neural network (CNN) is used to build a deep learning network model. There are three types of neural networks, namely ALexNet, VGG19 and ResNet-50.
The accuracy, sensitivity, specificity, positive predictive value and negative predictive value of the preliminary results all exceeded 0.90. The parameters of the neural network are fine-tuned after being calculated by the GPU.
According to the experimental results, the presented methods could be used to distinguish between diffuse and localized, hydrops and pneumonia, hydrops and infiltration, masses and nodules in X-ray images. Meanwhile, the proposed algorithms, convolutional neural network model, might be applied to the feasibility of clinical image classification.

並列關鍵字

Artificial Intelligence ； Deep Learning Algorithm ； Convolutional Neural Network ； Classification ； Chest X-ray

參考文獻

[1] Baltruschat IM, Nickisch H, Grass M, Knopp T, Saalbach A. Comparison of Deep Learning Approaches for Multi-Label Chest X-Ray Classification. Sci Rep. 2019;9(1):6381.

[2] Liang G, Zheng L. A transfer learning method with deep residual network for pediatric pneumonia diagnosis. Comput Methods Programs Biomed. 2020;187:104964.

[3] Kumar K. Objective Image Quality Metrics for Ultrasound Imaging, Proceedings of ISSN: 2012; 2278-0181.

[4] Fukushima, K. Fukushima K. Neocognitron: a self organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biol Cybern. 1980;36(4):193-202..

[5] 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, 1998;2278-2324.