ADHD based on deep network and subtype diagnosis method thereof
By analyzing video sequences using the Cognition_Net deep neural network and extracting physiologically specific indicators, the accuracy and consistency issues of existing ADHD diagnostic methods are resolved, achieving efficient diagnosis of ADHD and its subtypes.
Patent Information
- Application Number
- CN202511798014.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-03-20
AI Technical Summary
Existing methods for diagnosing ADHD rely on subjective approaches and lack clear biomarkers. Furthermore, existing objective methods such as fMRI, EEG, and eye-tracking devices suffer from complex data processing, limited accuracy, or susceptibility to interference, making it difficult to achieve high accuracy and consistency in diagnosis.
By employing the Cognition_Net deep neural network, physiologically specific indicators such as attention, average eye movement velocity, and average pupil size change velocity are extracted from short-term natural video sequences. A lightweight gaze-tracking network is constructed, and combined with multi-scale feature fusion and wavelet attention feature fusion modules, high-precision diagnosis of ADHD and its subtypes is achieved.
Without the need for specialized equipment or specific conditions, it achieves a 99% accuracy rate in distinguishing children with ADHD from healthy children, and can differentiate between the three subtypes of ADHD, with performance approaching that of professional-grade eye trackers.