[1M4-CC] Late Breaking Research Session - A

[1M4-CC-05] Can deep-learning predict pediatric brain age from CT images?

Ren Morita¹, Saya Ando², Daisuke Fujita¹, Manabu Nii¹, Kumiko Ando³, Reiichi Ishikura³, Syoji Kobashi¹ (1. Graduate School Engineering, University of Hyogo, 2. Hyogo Prefectural Amagasaki General Medical Center, 3. Kobe City Medical Center General Hospital)

One of the indicators for diagnosing brain diseases in children using brain imaging is to evaluate the progress of normal brain development in underdeveloped and premature infants. However, there is no quantitative method to estimate the degree of brain development, and diagnosis is currently based on the experience of doctors. Therefore, the lack of doctors who can diagnose and the lack of quantitative methods are problems. In this study, we propose a method to predict the age of brain development from pediatric brain CT images. By quantitatively evaluating the progress of brain growth in pediatric, we aim to clarify growth disorders and prematurity, and to support brain diagnosis by suggesting possible brain diseases. The proposed method consists of two major steps. First, we propose a method to extract head regions from 3D CT images. The pediatric brain CT images includes the hands and fingers of adults in the image because pediatric have difficulty sitting still during CT examinations. By removing them, we limit the region of interest to the head region, which is the target of brain age predicts. Next, we propose a new network model that extracts features from CT images using a 3D convolutional neural network (3D CNN) to predict the age of brain development in all the connectivity layers. The performance of this model was evaluated using 60 neurologically normal children between the ages of 0 and 3 years. The results showed that the root mean square error between predicted and actual age was 7.80 (months), and the correlation coefficient was 0.801.