Task Definition:
Interpretation reasoning in radiology report generation
Interpretation reasoning in radiology report generation
This task aims to generate interpretive reports from chest radiography.
Input
Chest radiography (image)
Interpretable report of diagnosis process (text)
The Interpretable report consists of Disease label (Diagnosis) and Diagnosis Information (Anatomical location, etc.)
Additional data
This task allows using any dataset, KB, etc.
Output
Interpretive report (text)
Report generated through the entire diagnosis process. For more information, please click here.
Procedures of Radiology Report
Patient Information (e.g., patient’s name, age, gender, …)
Examination Details (e.g., X-ray, MRI, CT scan, …)
Clinical History (patient’s clinical history and symptoms)
Technique (a description of how the imaging was performed)
Findings (observations from the images, including any abnormalities or changes detected)
Interpretation (The radiologist's interpretation of the findings)
Impression (A concise summary of the report, highlighting the most significant findings and their potential implications.)
Recommendations (or conclusion)
There are eight major procedures for writing a radiology report. This work focuses, among other things, on “interpretation”.
Because traditional CAD does not rely heavily on interpretation of the text inside. This is a characteristic of CAD, making diagnostic interpretation impossible and limiting its role. This work is an explainable AI study aimed at addressing the limitations of CAD.
Expected benefits of our task
Contribution to explainable CAD technology.
Contribution to causal interpretation techniques in chest radiology report generation.
Contribution to automatic annotation technology for generating datasets for causal interpretation of chest radiology images.
Contribution to a technology that achieves equivalent performance with fewer resources than GPT.