Open Access

ARTICLE

A Deep Learning-Based Automated Approach of Schizophrenia Detection from Facial Micro-Expressions

by Anum Saher¹, Ghulam Gilanie^1,*, Sana Cheema¹, Akkasha Latif¹, Syeda Naila Batool¹, Hafeez Ullah²

1 Department of Artificial Intelligence, Faculty of Computing, The Islamia University of Bahawalpur, Bahawalpur, 63100, Pakistan
2 Biophotonics Imaging Techniques Laboratory, Institute of Physics, The Islamia University of Bahawalpur, Bahawalpur, 63100, Pakistan

* Corresponding Author: Ghulam Gilanie. Email:

Intelligent Automation & Soft Computing 2024, 39(6), 1053-1071. https://doi.org/10.32604/iasc.2024.057047

Received 06 August 2024; Accepted 22 November 2024; Issue published 30 December 2024

Abstract

Schizophrenia is a severe mental illness responsible for many of the world’s disabilities. It significantly impacts human society; thus, rapid, and efficient identification is required. This research aims to diagnose schizophrenia directly from a high-resolution camera, which can capture the subtle micro facial expressions that are difficult to spot with the help of the naked eye. In a clinical study by a team of experts at Bahawal Victoria Hospital (BVH), Bahawalpur, Pakistan, there were 300 people with schizophrenia and 299 healthy subjects. Videos of these participants have been captured and converted into their frames using the OpenFace tool. Additionally, pose, gaze, Action Units (AUs), and land-marked features have been extracted in the Comma Separated Values (CSV) file. Aligned faces have been used to detect schizophrenia by the proposed and the pre-trained Convolutional Neural Network (CNN) models, i.e., VGG16, Mobile Net, Efficient Net, Google Net, and ResNet50. Moreover, Vision transformer, Swim transformer, big transformer, and vision transformer without attention have also been used to train the models on customized dataset. CSV files have been used to train a model using logistic regression, decision trees, random forest, gradient boosting, and support vector machine classifiers. Moreover, the parameters of the proposed CNN architecture have been optimized using the Particle Swarm Optimization algorithm. The experimental results showed a validation accuracy of 99.6% for the proposed CNN model. The results demonstrated that the reported method is superior to the previous methodologies. The model can be deployed in a real-time environment.

---

Keywords

---