Deep multi-kernel auto-encoder network for clustering brain functional connectivity data.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Data Science

Background:

• Brain diseases pose significant diagnostic challenges.
• Accurate clustering of functional connectivity data is crucial for understanding brain disorders.
• Existing clustering algorithms may not fully capture complex patterns in brain connectivity.

Purpose of the Study:

• To propose a novel deep-learning model, the deep multi-kernel auto-encoder clustering network (DMACN), for clustering brain functional connectivity data.
• To develop an end-to-end unsupervised clustering algorithm capable of learning advanced features for disease categorization.
• To evaluate the performance of DMACN against existing methods for brain disease classification.

Main Methods:

• Developed the deep multi-kernel auto-encoder clustering network (DMACN), an end-to-end deep learning model.
• Incorporated a self-expression layer with a kernel matrix for effective feature extraction.
• Proposed a novel loss function to enhance clustering performance during network training.
• Applied standard back-propagation for feature learning optimized for brain functional connectivity data.

Main Results:

• DMACN demonstrated superior performance in clustering brain functional connectivity data across multiple datasets.
• The algorithm achieved favorable results compared to existing deep auto-encoder clustering and other relevant clustering algorithms.
• DMACN successfully learned advanced features beneficial for distinguishing brain disease categories.

Conclusions:

• The deep multi-kernel auto-encoder clustering network (DMACN) is a promising deep-learning approach for unsupervised clustering of brain functional connectivity data.
• DMACN shows significant potential for the accurate and efficient recognition of brain diseases.
• This deep-learning-based clustering method offers a powerful tool for advancing brain disorder research and diagnostics.