Answers are grounded in this paper and cite their sources.
Intelligent systems are widely and increasingly used in medicine today. A particularly urgent task is to develop computer-aided diagnosis (CAD) systems, a tool that improves the performance of medical professionals in the context of the constantly increasing amount of medical data to analyze. One of the important components of modern CAD systems is a recognition module that helps identify pathological processes in medical images. This study considers the task of training a convolutional neural network to recognize cerebral vascular aneurysms. It offers a fully convolutional UNet-based architecture; a method of data pre-processing; and a method of seamless prediction, which engages the division of the original image into a set of overlapping image patches. The study explores how the size of image patches may affect the training performance of a neural network. The results of conducted computational experiments, after statistical analysis, allow us to conclude that the patch size is not a determining parameter, since an increased image patch size does not improve recognition accuracy. Upon that, experiments showed substantial growth in recognition accuracy, when a batch of training samples is expanded with all other parameters being fixed.
No takes yet. Share an insight, caveat, or question.
Kruzhalov et al. (2024) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: