東北大学 大学院情報科学研究科 情報基礎科学専攻 計算機構論分野
(東北大学 工学部 電気情報物理工学科 情報工学コース)
青木・伊藤(康)研究室

Age Estimation from Brain MAI Images Using Modified Squeeze-and-Excitation Transformer

Rizuki Oura (Tohoku University) , Koichi Ito (Tohoku University) , Takafumi Aoki (Tohoku University)
The International Forum on Medical Imaging in Asia, January 2026.
Abstract

The human brain has presented age-related morphological changes, which enables the estimation of a subject’s age from a brain magnetic resonance (MR) image. Recently, various CNN-based age estimation methods from MR images have been proposed. Among them, Squeeze-and-Excitation Transformer (SQET)1) is known to extract long-range local features, while does not consider the global morphological differences, such as those related to gender. We propose an age estimation method that integrates gender information into the SQET framework. Specifically, we extend the final fully connected layer in SQET and estimate age by concatenating gender information to the image features. We also incorporate data augmentation to enhance model generalization and a ranking loss to enforce the correct order of estimated ages.

戻る