Deep Learning Approaches for Mental Health Detection Using Facial Emotion Recognition: A Systematic Review

Abstract

: Depression, anxiety, stress have been on a rise across the world, and the need for early, accurate, and scalable diagnosis solutions is growing. The conventional assessment techniques are subjective, time-consuming and rely on clinical experience. Facial Emotion Recognition (FER) has been a promising non-invasive method for identifying mental health disorders using AI technology of facial expression analysis in recent years. This paper systematically reviews 50 research papers published from 2021 to 2024, which specifically concentrate on mental health detection based on FER using deep learning. The review covers different deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks and hybrid CNN-LSTM models. It also summarizes recent research on preprocessing methods, datasets, evaluation metrics, and multimodal learning methods. The results show that CNN-LSTM and attention based models have higher accuracy, by learning both spatial and temporal emotional features each. Moreover, multimodal integration of speech, physiological signals and facial expression contributes to the robustness and reliability of the system. Even with these progressions, there are still a number of hurdles to overcome, such as data imbalance, cultural bias, overfitting, and few data validations in real-world scenarios. Further, the deployment of real-time applications is limited by the computational complexity and by the need of lightweight models. In this review, we point out some major research gaps and suggest future directions, including explainable AI models, diverse and clinically validated datasets, and efficient architectures for real-time applications. The study is overall useful to researchers and practitioners for creating reliable and scalable systems for detecting mental health problems by using FER.