In clinical practice, the most frequently performed imaging test is CT, with chest CT being widely used. The purpose of chest CT is to check for abnormalities in the lungs and surrounding structures. Although the thyroid, located in the neck, may be visible on CT scans, thyroid nodules or thyroid cancer are often missed due to a lack of interpretation by radiologists. To address this, an artificial intelligence (AI) solution was developed to assist radiologists in identifying thyroid nodules.
The system extracts chest CT images from the hospital's Picture Archiving and Communication System (PACS) and builds a dataset on the thyroid and thyroid nodules. Clinical information is extracted from the hospital's Electronic Health Record (EHR) system, and AI software was developed to identify nodules within the thyroid through labeled data. A convolutional neural network (CNN)-based AI model was fine-tuned to improve performance, and the development was completed after testing.
The AI model was applied to the backup system of the radiology department's PACS at a clinical testing hospital, where its ability to detect thyroid nodules in chest CT images was evaluated. Performance was assessed based on the radiologists' interpretations, and the model was certified by the Ministry of Food and Drug Safety (Certification No. 23-4085) after a review of clinical trial data and technical documents.
This solution uses chest CT scans, the most frequently performed test in hospitals and medical institutions, to provide additional information to radiologists and clinicians by identifying thyroid abnormalities during lung disease diagnoses. In cases where low-dose CT is performed, the system helps detect thyroid abnormalities and recommends further thyroid ultrasound for confirmation of thyroid nodules.