Enhancing Radiology Clinical Histories Through Transformer-Based Automated Clinical Note Summarization.

Area of Science:

• Radiology
• Medical Informatics
• Natural Language Processing

Background:

• Radiology requests often lack sufficient clinical information.
• Electronic Health Records (EHRs) are cumbersome, hindering data extraction.
• Efficiently accessing clinical data is crucial for accurate radiology reports.

Purpose of the Study:

• To develop an automated tool for clinical note summarization.
• To extract pertinent clinical information for radiological assessments.
• To enhance radiologist workflow efficiency.

Main Methods:

• Utilized a transfer learning methodology from Natural Language Processing (NLP).
• Fine-tuned a transformer model using a two-stage approach: denoising and summarization.
• Trained on 1000 clinical notes from knee MRI studies, manually summarized by radiologists.

Main Results:

• The model achieved a 97% text condensation of clinical notes.
• Summaries showed high alignment with radiologist summaries (0.9 cosine similarity, 40.18 ROUGE-1 score).
• Demonstrated fair agreement (0.32 Fleiss kappa) on improved clinical history relevance.

Conclusions:

• The automated tool effectively summarizes clinical notes for knee MRI studies.
• This technology has the potential to improve radiology reporting efficiency and accuracy.
• Automated summarization can streamline the extraction of critical clinical information for radiologists.