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

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