Automated Detection of Invasive Fungal Infections in Clinical Reports Using Medical Language Models.

Area of Science:

• Medical Informatics
• Computational Linguistics
• Infectious Diseases

Background:

• Invasive fungal infections (IFIs) present a severe threat to immunocompromised patients.
• Early detection of IFIs is crucial for effective treatment and improved patient outcomes.
• Current methods for IFI detection from clinical text may be suboptimal.

Purpose of the Study:

• To evaluate the efficacy of advanced Natural Language Processing (NLP) techniques for detecting IFIs in clinical reports.
• To compare the performance of transformer-based pre-trained language models (PLMs) and generative large language models (LLMs) against existing methods.
• To explore the benefits of a hybrid NLP approach for IFI identification.

Main Methods:

• Utilized transformer-based pre-trained language models (PLMs) for IFI detection.
• Employed generative large language models (LLMs) in the IFI detection pipeline.
• Developed and tested a hybrid NLP approach combining diverse models.
• Evaluated performance on the public CHIFIR benchmark dataset.

Main Results:

• Advanced NLP methods, including PLMs and LLMs, demonstrated superior performance in IFI detection compared to prior techniques.
• A hybrid NLP approach achieved high accuracy, missing only one positive case on the CHIFIR dataset.
• The study confirmed the effectiveness of modern NLP in analyzing clinical text for disease detection.

Conclusions:

• Modern NLP techniques, particularly PLMs and LLMs, offer significant value for improving the detection of invasive fungal infections.
• Combining diverse NLP approaches enhances detection accuracy and robustness.
• These findings support the integration of advanced NLP tools into clinical workflows for better IFI surveillance.