Automatic Medical Code Assignment via Deep Learning Approach for Intelligent Healthcare.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Natural Language Processing for Clinical Text

Background:

• Healthcare 4.0 generates vast amounts of diverse patient data, including unstructured medical texts.
• Medical records are crucial for clinical insights but challenging to process due to their free structure and variability.
• Manual International Classification of Diseases (ICD) coding is essential for clinical and financial decisions but is labor-intensive, costly, and prone to errors.

Purpose of the Study:

• To develop an automated system for predicting ICD codes from unstructured medical text.
• To improve the efficiency and accuracy of medical coding processes.
• To facilitate the secondary use of clinical data for informatics.

Main Methods:

• A deep learning approach combined with a medical topic mining method was employed.
• The system was trained and evaluated on the Medical Information Mart for Intensive Care (MIMIC-III) dataset.
• Performance was also assessed on in-house ICD-10 datasets for specific conditions like atrial fibrillation.

Main Results:

• The proposed system achieved a 5% increase in F1 score on the MIMIC-III dataset compared to state-of-the-art methods.
• High F1 scores were obtained for atrial fibrillation: 96% on in-house ICD-10 datasets and 93.3% on MIMIC-III.
• The system demonstrated suitability for multiple ICD versions and languages.

Conclusions:

• The developed Artificial Intelligence-based coding system significantly enhances the efficiency and accuracy of human coders.
• This automated approach accelerates the secondary use of clinical data for advanced informatics applications.
• The system offers a scalable and effective solution for the challenges of medical text processing and ICD coding.