Depressive Disorder Remote Detection through Touchscreen Typing Behaviour.

Area of Science:

• Digital biomarkers
• Mental health informatics
• Machine learning in healthcare

Background:

• Depressive Disorder (DD) is a major global health concern and cause of disability.
• Objective and passive screening tools are needed for early detection and management of DD.
• Individual kinetic expressions, including smartphone usage patterns, offer insights into mental status.

Purpose of the Study:

• To investigate the utility of keystroke dynamics from touchscreen typing for detecting depressive disorders.
• To evaluate deep learning models for identifying depressive tendencies based on digital biomarkers.
• To compare the performance of Convolutional Neural Networks (CNN), Long-Short-Term-Memory (LSTM), and CNN-LSTM models.

Main Methods:

• Collected 23,264 typing sessions from 10 DD patients and 14 healthy controls (HC).
• Utilized keystroke sequences captured unobtrusively during routine smartphone interaction.
• Applied and compared CNN, LSTM, and CNN-LSTM models using two feature combinations and leave-one-subject-out (LOSO) cross-validation.

Main Results:

• The LSTM-with-hold-time (LSTM-HT) model achieved the highest performance.
• An Area Under Curve (AUC) of 0.86 was obtained for the best model.
• High sensitivity (0.8) and specificity (0.93) were reported, indicating strong diagnostic capability.

Conclusions:

• Keystroke dynamics serve as a viable digital biomarker for screening depressive disorders.
• Deep learning approaches, especially LSTM, can effectively detect depressive tendencies from typing patterns.
• Findings support the development of objective digital tools for real-world mental health screening and monitoring.