Automated image label extraction from radiology reports - A review

Area of Science:

• Medical Imaging
• Natural Language Processing (NLP)
• Machine Learning

Background:

• Machine learning models require extensive annotated data for training, which is scarce in medical imaging due to the need for expert physician annotation.
• Manual labeling of medical images by physicians is time-consuming and costly, hindering the development of large-scale datasets.
• Natural Language Processing (NLP) offers a potential solution for automating the extraction of labels from radiology reports, thereby reducing annotation efforts.

Approach:

• This study presents a meta-analysis and a qualitative/quantitative systematization of literature from 2013-2023 on using NLP for medical image labeling.
• Four categories of NLP approaches were identified: symbolic, statistical, neural, and hybrid/comparative systems.
• The review synthesizes findings to provide a comprehensive overview of the field's progress and identify areas for future research.

Key Points:

• NLP tools can automatically extract labels from radiology reports, significantly decreasing the manual annotation workload for medical imaging datasets.
• The literature review identified four main types of NLP systems: symbolic, statistical, neural, and combined approaches.
• Despite advancements, current NLP techniques for radiology report analysis still present opportunities for further refinement and innovation.

Conclusions:

• Automated label extraction from radiology reports using NLP is a viable method to accelerate the creation of labeled medical imaging datasets.
• The systematic review highlights the evolution and diversity of NLP techniques applied to this domain.
• Further research and development in NLP are crucial for improving the accuracy and efficiency of automated medical image annotation.