A method for extracting tumor events from clinical CT examination reports

Area of Science:

• Medical Informatics
• Natural Language Processing
• Artificial Intelligence

Background:

• Accurate extraction of disease information from medical reports is vital for diagnosis and treatment.
• Clinical examination reports contain crucial patient health data.
• Structured data organization enhances medical data review and analysis.

Purpose of the Study:

• To introduce a new technique for medical event extraction (EE) from unstructured clinical text examination reports.
• To improve the accuracy and efficiency of extracting key disease-related information.
• To leverage Machine Reading Comprehension (MRC) for enhanced medical data analysis.

Main Methods:

• Developed a two-sub-task approach: Question Answerability Judgment (QAJ) and Span Selection (SS).
• Utilized BERT for a question answerability discriminator (Judger) to filter unanswerable questions.
• Employed attention mechanisms and a bidirectional LSTM (BiLSTM) for precise answer span prediction.

Main Results:

• The proposed method achieved state-of-the-art results in medical event extraction.
• Demonstrated strong word representation capabilities and effective contextual information extraction.
• Achieved a notable F1 score, outperforming existing methods.

Conclusions:

• The new MRC-based technique significantly enhances the extraction of medical events from reports.
• The model's ability to identify answerable questions and precise answer spans improves clinical data utility.
• This approach offers a powerful tool for better patient care through improved medical record analysis.