A machine learning methodology for medical imaging anonymization.

Area of Science:

• Medical Informatics
• Computer Vision
• Machine Learning

Background:

• Electronic Health Record (EHR) systems require robust privacy protection, especially in collaborative environments involving data sharing.
• De-identifying textual data is straightforward, but medical images present unique challenges for patient privacy.
• Sensitive information within medical images necessitates advanced de-identification techniques.

Purpose of the Study:

• To develop and evaluate a system for sensitive word identification in medical images.
• To enhance patient data privacy in collaborative medical settings.
• To facilitate secure exchange of anonymized patient data between institutions.

Main Methods:

• Implementation of a solution combining two machine learning models for sensitive word identification.
• Evaluation of the system's performance by three independent medical experts.
• Categorization of results into true positives (sensitive words removed), false negatives (identity retained), and false positives (data mistakenly removed).

Main Results:

• The developed system achieved a high F1-score of 0.94 for sensitive word identification.
• Expert evaluation confirmed the system's effectiveness in removing sensitive information.
• The methodology demonstrated a low rate of false negatives and false positives.

Conclusions:

• The proposed machine learning approach is highly effective for sensitive word identification in medical images.
• The system significantly contributes to protecting patient privacy in collaborative medical environments.
• This tool is relevant for improving the secure exchange of anonymized patient data across healthcare institutions.