Improving the Prescription Process Information Support with Structured Medical Prospectuses Using Neural Networks

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Natural Language Processing

Background:

• Effective patient treatment requires comprehensive information on patient status and suitable medications.
• Medical prospectuses contain crucial drug information but are often unstructured.
• Automating the extraction and structuring of this information is essential for clinical decision support.

Purpose of the Study:

• To develop and evaluate three neural network methods for structuring medical prospectus sections.
• To improve the accuracy and efficiency of information extraction from medical documents.
• To facilitate better treatment decisions by organizing key data from drug information leaflets.

Main Methods:

• Utilized three distinct neural network architectures for text structuring.
• Trained networks using structured data from three web sources, including prospectuses with uniformized section names.
• Conducted tests on Romanian medical prospectuses, comparing accuracy and execution time.

Main Results:

• Convolutional neural networks demonstrated higher accuracy in structuring prospectus sections.
• Uniformized section names led to improved accuracy compared to original prospectus names.
• The developed methods provide structured data for decision support applications.

Conclusions:

• Neural network-based structuring of medical prospectuses enhances information accessibility.
• Convolutional networks and standardized section naming are key to maximizing accuracy.
• This approach supports clinical decision-making by efficiently matching treatments to patient conditions.