Researchers at Kaunas University of Technology (KTU) have developed an innovative AI model that identifies depression by analyzing both speech and brain neural activity.

With depression affecting approximately 280 million people globally, this research aims to address a significant mental health challenge.

The new AI model boasts an impressive accuracy rate of 97.53% in diagnosing depression, significantly outperforming existing diagnostic techniques.

EEG data for the study was sourced from the Multimodal Open Dataset for Mental Disorder Analysis (MODMA), with recordings taken while participants were awake and relaxed.

Musyyab Yousufi, a Ph.D. student at KTU, highlights that while facial expressions can indicate emotional states, voice data is a more reliable source due to its subtlety.

As the model evolves, it must provide explanations for its diagnostic decisions, aligning with the growing demand for explainable AI (XAI) in healthcare.

The research findings were published in the Brain Sciences Journal under the title "Multimodal Fusion of EEG and Audio Spectrogram for Major Depressive Disorder Recognition Using Modified DenseNet121."

Utilizing a multimodal approach, the model integrates speech and neural data, allowing for a more precise analysis of emotional states compared to traditional single-data methods.

The algorithm employs a modified DenseNet-121 deep-learning model, transforming collected EEG and audio signals into spectrograms for effective depression detection.

To enhance the model's reliability, participants engaged in activities involving natural language use, such as answering questions and describing images, during audio data collection.

However, the model's effectiveness will require further clinical trials and refinement, particularly to address challenges related to insufficient data due to privacy concerns.

Rytis Maskeliūnas, a KTU professor involved in the project, emphasizes the need for more objective diagnostic methods to improve accessibility for individuals suffering from depression.