DRCNN_TLe: A deep recurrent convolutional neural network with transfer learning through pre-trained embeddings for...

Area of Science:

• Medical Informatics
• Health Informatics
• Computational Medicine

Background:

• The International Classification of Diseases (ICD) is a global standard for health data, yet its adoption and application in Pakistan, particularly for liver disease epidemiology, are limited.
• Automated ICD coding systems are not widely implemented, hindering efficient health information management and research in the region.

Purpose of the Study:

• To annotate ICD codes for a liver transplant database (MIMLT) at a Pakistani medical institute.
• To determine the spectrum of liver disease burden and identify common indications for liver transplantation.
• To develop and implement an automated ICD coding system using a novel deep learning model.

Main Methods:

• ICD codes were manually annotated for the Medical Information Mart for Liver Transplantation (MIMLT) database.
• The frequency of ICD codes and the spectrum of liver disease indications for transplantation were analyzed.
• A Deep Recurrent Convolutional Neural Network with Transfer Learning through pre-trained Embeddings (DRCNN_TLe) model was developed for automated ICD coding.

Main Results:

• The MIMLT database contained 34 unique ICD codes, with V70.8 being the most frequent.
• Chronic hepatitis C (ICD-10 code 070.54) was identified as the leading indication for liver transplantation.
• The DRCNN_TLe model demonstrated improved performance using pre-trained word embeddings on a small dataset, achieving state-of-the-art results.

Conclusions:

• Automated ICD coding using the DRCNN_TLe model is feasible and effective, even with limited domain-specific data, by leveraging pre-trained embeddings.
• This approach can significantly enhance the analysis of liver disease burden and support clinical decision-making in regions with low ICD adoption.
• The study provides a foundation for implementing advanced health informatics tools in Pakistan's healthcare system.