Preliminary Text Analysis from Medical Records for TB Diagnosis Support

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Natural Language Processing

Background:

• Tuberculosis (TB) is a major global health threat and a leading cause of death.
• Artificial intelligence (AI) decision support systems offer benefits in healthcare, particularly for diagnosis and screening.
• Data availability is a significant challenge for developing AI in healthcare, especially in developing countries.

Purpose of the Study:

• To construct a novel dataset from medical records of patients suspected of tuberculosis.
• To explore the content of this dataset for potential use in AI-driven diagnostic tools.
• To present a strategy for leveraging natural language processing (NLP) to support tuberculosis diagnosis in data-limited settings.

Main Methods:

• Dataset construction using de-identified medical records of individuals with suspected tuberculosis.
• Exploratory data analysis of the constructed dataset's content and characteristics.
• Proposal of an NLP-based approach for subsequent analysis to support TB diagnosis.

Main Results:

• A new dataset of medical records for tuberculosis screening has been successfully created.
• Initial exploration reveals the dataset's suitability for NLP applications.
• The methodology provides a viable path for developing AI diagnostic support.

Conclusions:

• The developed dataset and proposed NLP strategy can facilitate AI-based tuberculosis diagnosis in resource-limited settings.
• This approach addresses the critical issue of data scarcity in developing countries.
• It offers a pathway to enhance healthcare support for tuberculosis through accessible medical data and AI.