Improving the Accuracy of a CNN Model by Preprocessing Input Images with Modified Filtering-masks

We propose a concise method to improve the inference accuracy of a convolutional neural network model for image classification. The characteristics of the input images are sharpened by a modified 5 ´ 5 mask before training and testing. The practice data were acquired from liver cancer MRI scanning at a collaborative hospital. We established the datasets using separated scanned images, which were labeled 1 or 0 to represent images with or without a cancer focal area, respectively. Scanned files from 45 patients were adopted for this study with each of them providing hundreds of separated images. We predicted one patient’s longitudinal cancer position in the liver to illustrate the merit of our approach. &nbsp