[Comparison of the Impact of Deep Learning Techniques on Low-noise Head CT Images]

This study aims to compare the effects of two types of deep learning (DL) techniques on brain CT values, image noise content, and contrast-to-noise ratio (CNR) between white and gray matter in low-noise head CT images, along with adaptive iterative dose reduction 3D (AIDR 3D).

Twenty-one normal patients with no abnormal findings who underwent head CT for identification of acute illness were included in the study. DL techniques used were Advanced intelligent Clear-IQ Engine (AiCE, Canon Medical systems, Tochigi, Japan) and PixelShine (FUJIFILM Medical, Tokyo, Japan). We performed CT value measurements of 26 cerebrum regions, image noise measurements, and CNR calculations. We also conducted a visual assessment of image noise and white matter-gray matter contrast on a 5-point scale.

Image noise significantly decreased with DL techniques. CT values changed significantly with AiCE. CNR for white matter-gray matter was the highest with PixelShine (P<0.01). The visual assessment of white matter-gray matter contrast was the highest for PixelShine and the lowest for AiCE (P<0.01).

While DL techniques reduce image noise, there are differences in CT values and visual impression, especially white matter-gray matter contrast, so care should be taken when using it.