Age Prediction from Baby Brain MRI Using Convolution-Based Architectures

Dataset

The UNC / UMN Baby Connectome Project, which includes 343 healthy term-born subjects. Images from subjects aged 0-2 years are selected, which contain 1355 T1 and T2 MRI images.

Result

The test performance of all four models under different input image and input mode based on MAE metric.

Input Mode	Middle Slices - T1	Middle Slices - T2	All Brain - T1	All Brain - T2
CNN-5	1.175	1.187	1.973	1.926
ASPP-CNN	1.4464	1.333	2.213	2.062
CBAM-CNN	1.396	1.404	2.132	2.205
ResNet	1.169	1.107	1.912	1.895