ICD Coding from Clinical Text Using Multi-Filter Residual Convolutional Neural Network.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Natural Language Processing

Background:

• Automated ICD coding is crucial for efficient medical billing, saving time and labor.
• Previous state-of-the-art models used a single convolutional layer, limiting their ability to represent diverse text fragments.
• Variations in text length and grammar pose challenges for fixed-length convolutional architectures in ICD coding.

Purpose of the Study:

• To develop an advanced deep learning model for automated ICD coding that addresses limitations of prior approaches.
• To improve the accuracy of International Classification of Disease (ICD) code assignment from clinical text.
• To introduce a novel Multi-Filter Residual Convolutional Neural Network (MultiResCNN) for enhanced document representation.

Main Methods:

• Proposed a Multi-Filter Residual Convolutional Neural Network (MultiResCNN) architecture.
• Utilized multi-filter convolutional layers to capture text patterns of varying lengths.
• Incorporated residual convolutional layers to expand the receptive field for comprehensive document understanding.
• Evaluated the model on the MIMIC-III and MIMIC-II datasets.

Main Results:

• The MultiResCNN model outperformed the state-of-the-art on 4 out of 6 metrics for the full MIMIC-III dataset.
• Achieved superior performance across all evaluation metrics on the top-50 MIMIC-III code set and the full MIMIC-II dataset.
• Demonstrated significant improvements in automated ICD coding accuracy compared to existing methods.

Conclusions:

• The proposed MultiResCNN is effective for automated ICD coding, outperforming previous state-of-the-art models.
• The model's architecture, featuring multi-filter and residual layers, enhances the representation of clinical text for accurate code assignment.
• This work provides a more robust deep learning solution for improving the efficiency and accuracy of medical billing through automated ICD coding.