Unsupervised information extraction from italian clinical records.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Data Mining

Background:

• Clinical records contain valuable information but are often unstructured.
• Extracting this data manually is time-consuming and prone to errors.
• Automated methods are needed to efficiently process clinical text.

Purpose of the Study:

• To apply an unsupervised text mining technique for information extraction from Italian clinical records.
• To explore relationships between domain entities identified in clinical text.
• To evaluate different clustering methods for relation discovery.

Main Methods:

• Utilized natural language processing (NLP) tools and a metathesaurus to extract domain entities.
• Employed clustering techniques to analyze relationships between extracted entity pairs.
• Conducted experiments on text from 57 medical records, analyzing over 20,000 potential relations.

Main Results:

• The unsupervised text mining approach successfully extracted domain entities from clinical records.
• Clustering analysis revealed relationships between entity pairs.
• Cosine similarity distance was identified as a more effective clustering metric compared to City Block or Hamming distances.

Conclusions:

• Unsupervised text mining is a viable method for information extraction from Italian clinical records.
• Cosine similarity is the preferred distance metric for clustering entity pairs in this context.
• Further research can refine these methods for broader clinical data analysis.