- Log in to post comments
We are absolutely thrilled to announce our successful presentation at the Metadata and Semantics Research (MTSR) Conference 2023. This achievement marks a significant milestone in our journey to support Southern African radiologists through the implementation of radiology report metadata.
Introduction
The research focuses on addressing the critical issue of a shortage of radiologists in Southern African countries, particularly in Zambia, where there are only nine radiologists serving a population of 19.6 million. It highlights the potential of Artificial Intelligence (AI) in enhancing radiology workflow and emphasizes the need to create a framework for effective radiology report generation.
Motivation and Research Problem
The shortage of radiologists in Southern Africa poses a significant challenge, with just nine radiologists serving Zambia's population of 19.6 million. The research recognizes the potential of AI-driven information systems to improve the efficiency of radiology workflow. It underlines the need for specialized medical training and human-in-the-loop report production. Moreover, the research highlights the importance of clarifying the clinical efficacy of AI-driven systems.
Research Objectives
The primary objective of this research is to develop algorithms and theories that will assist radiologists in their practice. This includes characterizing effective radiology reports, modeling their structure, and creating tooling to support human-in-the-loop authoring.
Existing Work
Existing work in this field reveals that there are no metadata standards or ontologies designed for the semantic annotation of radiology reports. While DICOM offers some Information Object Definitions, they do not cover all aspects comprehensively. Furthermore, current report generation methods using deep learning do not include specific terminology or metadata standards for labeling reports.
Terminology Development
The research involved the meticulous process of gathering radiology report templates from various sources. This effort resulted in the identification of 3199 terms from Radiological Society of North America (RSNA) templates and 323 terms from recommendations and publications. These terms were further aligned with SNOMED CT to create a comprehensive radiology terminology.
Utility and Use Case
The development of this comprehensive radiology terminology serves a practical purpose. It plays a vital role in designing Natural Language Generation (NLG) systems for generating high-quality reports. Radiologists often rely on their knowledge from past training institutions, resulting in variations in terminology. The terminology created in this research can be invaluable in producing context-sensitive structured report templates suitable for specific regions and preferences.
Conclusions and Future Work
In conclusion, this research has successfully created the first comprehensive and unbiased radiology terminology. To further enhance this field, future research will involve conducting mixed methods research with a broader group of radiologists and annotating existing Zambian radiology reports. This will help identify characteristics and areas for improvement, ultimately contributing to the advancement of radiology practices in Southern Africa.