NATURAL LANGUAGE PROCESSING BASED MACHINE LEARNING MODEL USING CARDIAC MRI REPORTS TO IDENTIFY HYPERTROPHIC...

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Hypertrophic Cardiomyopathy (HCM) is a prevalent genetic heart condition linked to sudden cardiac death in young adults.
• Accurate identification of HCM cases from electronic health records (EHR) is crucial for improved patient management and outcomes.
• Cardiac Magnetic Resonance (CMR) imaging provides valuable diagnostic data, including measurements and descriptive narratives in its reports.

Purpose of the Study:

• To develop an interpretable Natural Language Processing (NLP) model to automatically identify Hypertrophic Cardiomyopathy (HCM) patients.
• To leverage impression sections of CMR reports for automated HCM case identification.
• To enhance the efficiency of HCM diagnosis and risk stratification using EHR data.

Main Methods:

• Utilized CMR reports from patients with suspected HCM (1995-2019).
• Classified patients into 'yes HCM', 'no HCM', and 'possible HCM' categories.
• Developed a Random Forest (RF) model incorporating CMR measurements and impression features.

Main Results:

• The RF model achieved 86% accuracy using 608 features and 85% accuracy using 30 features.
• Demonstrated the effectiveness of NLP and ML in identifying HCM patients from CMR report text.
• Highlighted the potential of impression features for accurate HCM classification.

Conclusions:

• NLP-based models applied to CMR reports show significant promise for automated HCM patient identification.
• This approach can facilitate more efficient clinical management and transform healthcare delivery for HCM patients.
• Automated identification of HCM using EHR data can lead to improved patient outcomes and reduced adverse events.