Applying Deep Learning Model to Predict Diagnosis Code of Medical Records

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Documentation

Background:

• International Classification of Diseases (ICD) codes are crucial for healthcare referencing and billing.
• Manual ICD coding from patient records is time-consuming and prone to errors.
• Deep learning (DL) offers potential solutions for automating and improving the accuracy of ICD coding.

Purpose of the Study:

• To develop and evaluate a deep learning model for predicting ICD-10 codes using clinical notes.
• To assess the model's performance across different hospital departments.
• To demonstrate the potential of AI in streamlining the clinical coding process.

Main Methods:

• Utilized a dataset of 21,953 medical records from a university hospital's outpatient department (2016).
• Applied natural language processing (NLP) techniques, including Word2Vector, for text data processing.
• Developed and trained a deep learning-based convolutional neural network (CNN) model.

Main Results:

• The CNN model achieved clinically acceptable performance across five departments (Precision: 0.53-0.96, Recall: 0.85-0.99, F-score: 0.65-0.98).
• The model demonstrated superior performance in cardiology (Precision: 0.95, Recall: 0.99, F-score: 0.98).
• Significant improvement in automated ICD-10 code prediction accuracy was observed.

Conclusions:

• The proposed CNN model effectively predicts ICD-10 codes from clinical notes.
• This AI-driven approach can significantly reduce manual coding workload for physicians.
• The system has the potential to enhance diagnostic accuracy and operational efficiency in healthcare settings.