Automatic colorization of non-enhanced brain CT images for clinical diagnosis

 The frequent use of brain computed tomography (CT) scans in emergency settings necessitates accurate reporting of CT results as quickly as possible. Conventional CT scans produce grayscale images, requiring window width and center level changes, resulting in a need for time-consuming interpretation by experienced radiologists. This study aimed to design a novel software application for automatic smart colorization of conventional brain CT images and to evaluate the diagnostic accuracy, visual quality, ease of diagnosis, and reporting time for color CT images compared to conventional grayscale CT images.

 First, we designed an application that converted non-enhanced grayscale brain CT images into color images according to the Hounsfield unit value of different tissues (e.g., brain, fat, bone, fluid, air) with minimal noise so that all brain tissues could be evaluated using one window level. This process took less than one second, without the need for high-end systems. Next, 75 printed images (25 unprocessed grayscale CT, 25 processed color CT, and 25 magnetic resonance imaging [MRI]) from 25 patients with hemorrhagic or ischemic stroke were read by two experienced radiologists. The radiologists scored the CT images from each patient (unprocessed grayscale and processed color) on a ten-point scale for visual quality and ease of diagnosis compared to the MRI image.

 The mean visual quality score was 18% higher and the mean ease of diagnosis score was 23% higher for colorized images than for grayscale images (both P < 0.001). Statistically, there were no significant differences in the diagnostic accuracy or reporting time between color and grayscale images.

 This is the first study to report automatic smart colorization of non-enhanced brain CT images, producing high-quality colorized images with better visual quality and ease of diagnosis compared to grayscale CT. This low-cost solution can be widely applied in clinical settings, regardless of minimal facility or resource availability.