Effect of the Pixel Interpolation Method for Downsampling Medical Images on Deep Learning Accuracy  

作　　者：Daisuke Hirahara Eichi Takaya Mizuki Kadowaki Yasuyuki Kobayashi Takuya Ueda Daisuke Hirahara;Eichi Takaya;Mizuki Kadowaki;Yasuyuki Kobayashi;Takuya Ueda(Department of AI Research Lab, Harada Academy, Kagoshima, Japan;Department of Clinical Imaging, Graduate School of Medicine, Tohoku University, Sendai, Japan;Department of Advanced Biomedical Imaging Informatics, St. Marianna University School of Medicine, Kawasaki, Japan;Tohoku University Hospital AI Lab, Sendai, Japan)

机构地区：[1]Department of AI Research Lab, Harada Academy, Kagoshima, Japan [2]Department of Clinical Imaging, Graduate School of Medicine, Tohoku University, Sendai, Japan [3]Department of Advanced Biomedical Imaging Informatics, St. Marianna University School of Medicine, Kawasaki, Japan [4]Tohoku University Hospital AI Lab, Sendai, Japan

出　　处：《Journal of Computer and Communications》2021年第11期150-156,共7页电脑和通信（英文）

摘　　要：<strong>Background:</strong> High-resolution medical images often need to be downsampled because of the memory limitations of the hardware used for machine learning. Although various image interpolation methods are applicable to downsampling, the effect of data preprocessing on the learning performance of convolutional neural networks (CNNs) has not been fully investigated. <strong>Methods:</strong> In this study, five different pixel interpolation algorithms (nearest neighbor, bilinear, Hamming window, bicubic, and Lanczos interpolation) were used for image downsampling to investigate their effects on the prediction accuracy of a CNN. Chest X-ray images from the NIH public dataset were examined by downsampling 10 patterns. <strong>Results:</strong> The accuracy improved with a decreasing image size, and the best accuracy was achieved at 64 × 64 pixels. Among the interpolation methods, bicubic interpolation obtained the highest accuracy, followed by the Hamming window.<strong>Background:</strong> High-resolution medical images often need to be downsampled because of the memory limitations of the hardware used for machine learning. Although various image interpolation methods are applicable to downsampling, the effect of data preprocessing on the learning performance of convolutional neural networks (CNNs) has not been fully investigated. <strong>Methods:</strong> In this study, five different pixel interpolation algorithms (nearest neighbor, bilinear, Hamming window, bicubic, and Lanczos interpolation) were used for image downsampling to investigate their effects on the prediction accuracy of a CNN. Chest X-ray images from the NIH public dataset were examined by downsampling 10 patterns. <strong>Results:</strong> The accuracy improved with a decreasing image size, and the best accuracy was achieved at 64 × 64 pixels. Among the interpolation methods, bicubic interpolation obtained the highest accuracy, followed by the Hamming window.

关 键 词：Downsampling INTERPOLATION Deep Learning Convolutional Neural Networks Medical Images Nearest Neighbor BILINEAR Hamming Window Bicubic LANCZOS 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]