# Application of multimodal data fusion and intelligent classification in medical coding with the MCoder-T model

## TL;DR

The MCoder-T model improves medical coding by combining text, images, and structured data, leading to better accuracy and automation.

## Contribution

The novel MCoder-T model introduces causal-to-mask attention mechanisms and multi-task learning for enhanced medical coding.

## Key findings

- MCoder-T outperforms traditional methods in case coding accuracy and productivity.
- The model integrates multimodal data effectively, showing reliable adaptability.
- It achieves a productivity improvement of 7% to 18% in medical coding tasks.

## Abstract

Assigning case codes is a complex problem in medical data processing, which includes multimodal data fusion and intellectual classification. Traditional case coding methods often have difficulties in managing different data sources and the complexity of the content of cases. This limits their effectiveness in actual application. To solve this problem, we propose the MCoder-T model, an intelligent case coding model, causal-to-mask attention mechanisms, integrated multimodal integration, and multi-task learning optimization. MCoder-T effectively improves case coding automation and classification accuracy by integrating text, medical images, and structured data. Experimental results show that the MCoder-T model outperforms traditional methods and other progressive models by several evaluation indicators, with an overall productivity improvement of 7% to 18%. The MCoder-T model enhances the automation of drop coding tasks and demonstrates reliable adaptability during multimodal data fusion and demonstrates broad application potential.