運用遷移學習技術的CheXNet建立胸部X光圖像之分類模型

Classification of Abnormality in Chest X-Ray Images by Transfer Learning of CheXNet

馬汪達(Mawanda Almuhayar)

指導教授 ： 盧鴻興 ； Nur Iriawan

國立交通大學/理學院/統計學研究所/碩士(2019年)

摘要

現今深度學習已引起了極大的關注，特別是醫學圖像的分類問題，它的效率與良好的表現足以與醫學圖像的專家相媲美。儘管已取得了這些成功，專家們仍堅信深度學習只對大型的數據集有效，使用於小型的數據集時，深度學習將產生出不良的表現。本研究旨在建構一個執行圖像分類的深度學習模型，對於資料量較少的胸部X光圖像資料集進行分類，並達到優秀的預測準確度。我們將透過使用公開的Shenzhen Hospital資料集，進行模型的建構與測試，對於正常、不正常的胸部X光圖像進行二元分類。通過使用基於不同的預處理與不同類型的學習技術所生成的不同類型的圖像作為導入，使模型能夠對於這個數據集執行準確的分類。最終結果表明，於裁剪後的資料集使用預先訓練的CheXNet模型與新訓練的全連接網路所建構的新模型，實現了最佳的預測結果。此外，模型的性能也受到圖像中特定區域的影響，如肺部外圍的其他區域、軀體外圍的黑色區域。

關鍵字

胸部X光 ； 深度學習 ； 遷移式學習 ； 分類 ； 異常 ； 預處理 ； CheXNet

並列摘要

Deep learning nowadays has attracted attention, especially in medical images classification because of its effectiveness and good performance that can compete with the medical images expert. Despite these successes there are the strong belief among experts that deep learning only efficient for the big datasets and for small datasets deep learning would produce a bad performance. For this study, it is aimed to build a deep learning model for image classification that can achieve high accuracy using chest x-ray images with relatively small dataset. We classify all normal chest x-ray images and all abnormalities in chest x-ray images into a binary classifier. We built and tested our model using the public dataset of Shenzen Hospital dataset. We use different type of input images based on different preprocessing and different type of learning technique so that the model can perform accurate classification for this particular dataset. Based on the result, pre-trained CheXNet with new trained fully connected network on cropped dataset can achieve the accuracy 0.8761, the sensitivity 0.8909, and the specificity 0.8621. The performance of the model also influenced by the certain area in the images, like other region outside the lung and black region outside the body.

並列關鍵字

chest x-ray ； deep learning ； transfer learning ； classification ； abnormalities ； preprocessing ； CheXNet

參考文獻

[1] P. Rajpurkar et. al., “Deep learning for chest radiograph diagnosis: A retrospective comparison of the CheXNeXt algorithm to practicing radiologists”, PLoS Med 2018;15(11):e1002686, 2018.

[2] S. Stirenko, et. al., “Chest X-ray analysis of tuberculosis by deep learning with segmentation and augmentation”, arXiv preprint arXiv:1803.01199, 2018.

[3] M. T. Islam, et. al., “Abnormality detection and localization in chest x-rays using deep convolutional neural networks,” arXiv preprint arXiv:1705.09850, 2017.

[4] P. Rajpurkar et. al., ‘‘CheXNet: Radiologist-level pneumonia detection on chest X-rays with deep learning,’’ arXiv preprint arXiv:1711.05225, 2017.

[5] G. Huang, et. al., “Densely connected convolutional networks.” arXiv preprint arXiv:1608.06993, 2016.