Self-Paced Learning and Privileged Information based Cascaded Multi-column Classification algorithm for ASD...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Autism spectrum disorder (ASD) is a significant childhood mental disorder.
• Computer-aided diagnosis (CAD) using resting-state functional magnetic resonance imaging (rs-fMRI) is a key research area for ASD.
• Existing methods like Learning Using Privileged Information (LUPI) often require multi-modality data, which is clinically challenging to acquire.

Purpose of the Study:

• To develop a novel, single-modal diagnostic algorithm for ASD.
• To adapt LUPI for clinical feasibility by avoiding additional imaging modalities.
• To improve the classification accuracy for ASD diagnosis.

Main Methods:

• Proposed a self-paced learning based cascaded multi-column Random Vector Function Link Network plus (SPL-cmcRVFL+) algorithm.
• Utilized rs-fMRI as single-modal data for initial classification model training.
• Employed self-paced learning (SPL) to adaptively select samples from simple to difficult based on loss values for training subsequent layers with privileged information.

Main Results:

• The SPL-cmcRVFL+ algorithm demonstrated accurate identification of individuals with ASD and normal controls.
• The proposed method achieved higher classification accuracy compared to other existing methods.
• The study validates the effectiveness of LUPI without multi-modality data.

Conclusions:

• The developed SPL-cmcRVFL+ algorithm offers a promising approach for accurate ASD diagnosis using single-modal rs-fMRI data.
• This method addresses the clinical challenge of acquiring multi-modality data.
• The findings suggest potential for improved early detection and intervention strategies for ASD.