Comparative Evaluation of Fixed Windowing Strategies on CT Brain Images Using Multiple Deep Learning Models

Abstract

Window setting in CT brain images is the crucial pre-processing step to examine the abnormalities for diagnosing disease. Recently, many methods have been proposed to determine the suitable window automatically instead of fixing the window. However, fixed windowing methods may still be used in clinical practice due to their simplicity and ease of use. Here, we propose to evaluate the 45 different fixed windowing in noncontrast cranial computer tomography (NCCT) images without computer tomography perfusion (CTp). The 15 latest deep learning models are performed on all interested windows to classify between the hyperacute or acute phases of ischemic stroke and normal brain. The experiments can provide the reference fixed windowing value to optimize the deep learning model and help clinicians to choose the appropriate fixed windowing values.