Detecting Ground Glass Opacity Features in Patients With Lung Cancer: Automated Extraction and Longitudinal Analysis...

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Oncology

Background:

• Ground-glass opacities (GGOs) on CT scans may indicate lung cancer, necessitating careful management.
• Electronic health records contain valuable GGO data, but it's often unstructured in clinical notes.

Purpose of the Study:

• To develop and validate a deep learning-based natural language processing (NLP) tool for automated extraction of GGO features.
• To utilize extracted features for understanding the longitudinal trajectory of GGO status from radiology notes.

Main Methods:

• A deep learning NLP pipeline using bidirectional LSTM and CRF was developed.
• The pipeline processed radiology notes from 13,216 lung cancer patients to extract GGO and granular features.
• Quality assessments and longitudinal analyses of nodule features (size, solidity) were performed.

Main Results:

• The NLP pipeline achieved high performance: 95-100% precision, 89-100% recall, and 92-100% F1-scores.
• The model extracted comprehensive GGO characteristics from 29,496 radiology notes of 4521 patients.
• Longitudinal analysis showed GGO size changes in 31.4% of patients, with 72.3% maintaining stable status and 23% progressing.

Conclusions:

• A deep learning NLP pipeline effectively extracts granular GGO features from electronic health records.
• This tool facilitates the study of GGO natural history, paving the way for improved lung cancer prevention and early detection.