A Deep Learning Framework for Automated ICD-10 Coding

Area of Science:

• Medical Informatics
• Computational Linguistics
• Artificial Intelligence in Healthcare

Background:

• The International Statistical Classification of Diseases and Related Health Problems (ICD) is crucial for coding diagnoses and procedures in Electronic Health Records (EHRs).
• Manual ICD coding is time-consuming and prone to errors, necessitating efficient automated solutions.
• Accurate ICD coding is vital for healthcare management, research, and billing.

Purpose of the Study:

• To develop and evaluate an automated coding system for assigning ICD codes to EHRs.
• To leverage Natural Language Processing (NLP) and Deep Learning (DL) for extracting relevant information from French medical texts.
• To assist physicians in the accurate and efficient assignment of diagnosis codes.

Main Methods:

• A pipeline combining NLP and DL models was developed to process French medical texts.
• The system was trained to extract key information and classify it into relevant ICD codes.
• The performance was evaluated on heterogeneous medical units, predicting 346 distinct diagnosis codes.

Main Results:

• The automated coding system achieved an average accuracy of 83% in predicting ICD diagnosis codes.
• The system demonstrated proficiency in handling diverse French medical text data.
• Physicians from the Medical Information Department (MID) validated the system's output.

Conclusions:

• The proposed NLP and DL pipeline offers an effective solution for automated ICD coding in EHRs.
• This approach can significantly improve the efficiency and accuracy of medical coding processes.
• The validated system shows promise for widespread adoption in clinical settings to support physicians.