Deep learning-based depression recognition through facial expression: A systematic review

Abstract

Depression is a type of prevalent mental illness that can lead to suicidal or self-harm behaviors in severe cases. Recently, depression recognition has garnered extensive attention from the deep learning community due to its urgent need to assist conventional diagnostic methods. Deep learning-based depression recognition through facial expression (DL-FEDR) is one of the most popular research directions. Therefore, this paper tries to summarize advances from 2017 to 2024 in DL-FEDR. We focus on (1) Various approaches of DL-FEDR are divided into two categories: spatial and spatial–temporal features. (2) These methods are analyzed from metrics results, ethical privacy, application scenarios and technological advancements. (3) Present challenges and future directions of DL-FEDR systems are discussed.