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

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Medical case coding is complex, involving multimodal data fusion and classification.
• Traditional methods struggle with diverse data sources and content complexity, limiting effectiveness.

Purpose of the Study:

• To introduce the MCoder-T model for intelligent medical case coding.
• To address limitations of traditional methods in handling multimodal data and complex case content.

Main Methods:

• Developed the MCoder-T model, incorporating causal-to-mask attention mechanisms.
• Integrated multimodal data (text, medical images, structured data) using multi-task learning optimization.

Main Results:

• MCoder-T significantly improves case coding automation and classification accuracy.
• Outperformed traditional and progressive models across multiple evaluation indicators.
• Achieved a 7% to 18% overall productivity improvement.

Conclusions:

• The MCoder-T model enhances automation for case coding tasks.
• Demonstrates reliable adaptability in multimodal data fusion.
• Shows broad application potential in medical data processing.