Using DICOM Metadata for Radiological Image Series Categorization: a Feasibility Study on Large Clinical Brain MRI...

Area of Science:

• Medical Imaging
• Machine Learning in Healthcare
• Radiology Informatics

Background:

• Machine learning (ML) holds promise for enhancing patient care, but clinical integration is hindered by infrastructure and process limitations.
• Automating the selection of relevant imaging data for ML algorithms is a significant challenge in clinical practice.
• Current systems lack efficient methods for identifying and routing specific medical imaging series for analysis.

Purpose of the Study:

• To develop and validate a methodology for the automated identification and routing of brain Magnetic Resonance Imaging (MRI) sequences.
• To address the challenge of selecting relevant inputs for image-related algorithms in clinical settings.
• To enhance the integration of machine learning tools into radiological workflows.

Main Methods:

• A novel methodology leveraging metadata from the Digital Imaging and Communications in Medicine (DICOM) standard was developed.
• The approach automates the identification of brain MRI sequences.
• The method was tested on two large-scale, multi-institutional brain MRI datasets comprising 40,000 studies.

Main Results:

• The proposed method demonstrated high efficiency, processing each series in less than 0.4 milliseconds.
• Accuracy rates for identifying relevant MRI sequences ranged from 97.4% to 99.96%.
• The approach showed excellent generalizability across different institutions and continents.

Conclusions:

• Automated identification of brain MRI sequences using DICOM metadata is feasible and highly accurate.
• This methodology significantly improves the efficiency and generalizability of routing imaging data for ML algorithms.
• The technique shows potential for broader application in other radiological imaging modalities.