Performance of Machine Learning Methods to Classify French Medical Publications

Area of Science:

• Computational linguistics
• Medical informatics
• Machine learning

Background:

• Medical documents are often in non-English languages, limiting their discoverability via standard English-based search engines like PubMed.
• Healthcare professionals require efficient access to medical information in their native language for effective patient care.
• Current search methods pose challenges for locating French-language medical literature.

Purpose of the Study:

• To develop and evaluate an automated Natural Language Processing (NLP) pipeline for categorizing French medical documents.
• To compare the performance of various machine learning (ML) and deep neural network (DNN) approaches for this task.
• To enhance the findability of French medical literature for healthcare professionals.

Main Methods:

• A Natural Language Processing (NLP) pipeline was designed for automatic categorization of French medical texts.
• Six different machine learning and deep neural network models were implemented and compared.
• The models were trained and tested on a large dataset of peer-reviewed Swiss medical journal articles published weekly over 15 years.

Main Results:

• The NLP pipeline achieved high accuracy in classifying medical documents by topic.
• An accuracy of 96% was reached for a 5-topic classification task.
• An accuracy of 81% was achieved for a more granular 20-topic classification task.

Conclusions:

• Automated NLP categorization is a viable and effective method for organizing French medical documents.
• Machine learning and deep learning models demonstrate strong performance in classifying medical topics.
• This approach significantly improves the accessibility and discoverability of crucial medical information in French.