Hidden-RAD:
Hidden Causality Inclusion in Radiology Report Generation
Hidden-RAD:
Hidden-RAD proposed the task of generating interpretative reports of chest radiographs. The goal of this task is “interpretative reasoning”, focusing on the diagnostic process of radiographs. This task provides participants with a report compiled following a diagnostic process. Through this work, we hope to improve AI that cannot be interpreted and expand the use of CAD (Computer-aided diagnosis) with a focus on diagnostic assistance.
What is interpretation?
Interpretation ultimately contributes to diagnosis through the process of accurately identifying normal structures and abnormalities in radiological images, understanding the features of abnormalities, and weighting these findings.
Limitations of previous research
Interpretation is not considered when generating reports.
This is a weakness in the functionality that the report provide to the doctor.
This weakness limits the utility of CAD.
A limitation of previous studies is that they do not take "interpretation" into account when generating reports. Because no one considers “Interpretation” in relevant datasets. So, this task expands the CAD (Explainable AI) by reasoning interpretation in the report along with data considering the “Interpretation”.