EMS: A Large-Scale Eye Movement Dataset, Benchmark, and New Model for Schizophrenia Recognition

Area of Science:

• Neuroscience
• Computer Science
• Psychiatry

Background:

• Schizophrenia (SZ) is a prevalent mental illness often accompanied by cognitive deficits.
• Eye-tracking technology offers a cost-effective method for characterizing cognitive impairments.

Purpose of the Study:

• To address the lack of large-scale, publicly available eye movement datasets and benchmarks for schizophrenia recognition.
• To introduce a novel deep learning model for improved eye movement-based schizophrenia detection.

Main Methods:

• A large-scale Eye Movement dataset for SZ recognition (EMS) was created, including data from 104 SZ patients and 104 healthy controls (HCs).
• A comprehensive benchmark comparing 13 existing psychosis recognition methods was conducted.
• A novel mean-shift-based network (MSNet) was proposed, integrating mean shift algorithm with convolution for feature extraction.

Main Results:

• The proposed MSNet demonstrated superior performance compared to existing methods in eye movement-based SZ recognition.
• The EMS dataset and benchmark results provide a valuable resource for the research community.
• MSNet effectively extracts subject features using a stimulus feature branch and a cluster center branch.

Conclusions:

• The EMS dataset and benchmark establish a foundation for future research in eye movement analysis for schizophrenia.
• The MSNet model offers a powerful baseline for advancing the study of eye movement-based schizophrenia recognition.
• Publicly releasing the dataset, benchmark results, and code facilitates further investigation and development in this field.