Inducing Vascular Grammars for Anomaly Classification in Brain Angiograms.

Area of Science:

• Medical Imaging
• Machine Learning
• Computational Biology

Background:

• Machine learning advances medical diagnostics, but lacks explainable heuristics for clinicians.
• Existing methods often require extensive training data, limiting their application in specific medical scenarios.

Purpose of the Study:

• To develop a method for medical condition assessment using human-observable structures like vascular formations.
• To leverage shape grammars for identifying diagnostic patterns in medical data with limited training cases.

Main Methods:

• Employed shape grammars, a rule-based technique, combined with recursive subgraph mining.
• Utilized segmented seven-tesla time-of-flight angiography MRI data (n=54) of human brain vasculature.
• Generated representative grammar rules and used their distribution for classification and anomaly detection.

Main Results:

• Successfully applied the method to human brain vasculature MRI data.
• Isolated ensembles of grammar rules that reliably implicated vascular conditions.
• Demonstrated the effectiveness of automated structured representations for analyzing biological form relationships.

Conclusions:

• Shape grammars are powerful tools for identifying indicative patterns in complex vascular networks.
• The proposed method enables effective medical condition assessment with limited data by using structured intermediary representations.
• This approach enhances the interpretability of machine learning in medical diagnostics.