Neural hypernetwork approach for pulmonary embolism diagnosis

Area of Science:

• Computational topology
• Network theory
• Machine learning

Background:

• Hypernetworks generalize network theory using topological simplicial complexes for many-body relations.
• Pulmonary embolism is a critical lung artery blockage with high fatality rates.

Purpose of the Study:

• To develop and validate a novel diagnostic approach for pulmonary embolism.
• To integrate Q-analysis with machine learning for analyzing incomplete medical datasets.

Main Methods:

• Utilized a Neural Hypernetwork based on Q-analysis and machine learning.
• Applied the method to a dataset of 1427 patients at risk for pulmonary embolism.
• The approach does not rely on imaging analysis for clinical parameters.

Main Results:

• The Neural Hypernetwork achieved 94% accuracy in recognizing patients who developed pulmonary embolism.
• This performance surpasses previous methods including statistical feature selection, partial least squares regression, and metric space topological data analysis.

Conclusions:

• A new integrative approach, the Neural Hypernetwork, was successfully developed for analyzing partial and incomplete datasets.
• This method offers a novel, non-imaging-based strategy for pulmonary embolism diagnosis.