AI Based Lesion-Detection Software Detects Incidental Lung Nodules on Chest X-Rays

In the field of radiology, artificial intelligence (AI) has made significant strides, particularly in the development of AI-based lesion-detection software for chest X-rays. These advancements have proven effective in real-world settings, including emergency departments, lung cancer screenings, and respiratory clinics. However, the impact of AI in identifying unexpected lung nodules in patients not initially presenting with chest-related issues has been less explored. Now, a new study has demonstrated that an AI-based lesion-detection software can be instrumental in daily medical practice, especially for spotting clinically significant incidental lung nodules in chest X-rays.

A group of researchers at Yonsei University College of Medicine (Gyeonggi-do, South Korea) used Insight CXR, v3 from Lunit (Seoul, South Korea) to evaluate how often clinically significant lung nodules were detected unexpectedly on chest X-rays and whether coexisting findings can aid in differential diagnoses. This software is intended to assist in the interpretation of both posterior-anterior and anterior-posterior chest X-rays. It is capable of detecting various lesions such as nodules, pneumothorax, consolidation, atelectasis, fibrosis, cardiomegaly, pleural effusion, and pneumoperitoneum. When a patient has a chest X-ray, the software automatically processes the image and adds a secondary file to the original image in the hospital’s Picture Archiving and Communication System (PACS). Clinicians can then consult the AI analysis, which is presented with a contour map, abbreviations, and an abnormality score.

In their study, the team reviewed the imaging results of 14,563 patients who had initial chest X-rays at outpatient clinics. Three radiologists classified nodules into four categories: malignancy (group A), active inflammation or infection requiring treatment (group B), postinflammatory sequelae (group C), and other conditions (group D). The software identified lesions when its abnormality score was above 15%. The findings revealed that the AI software unexpectedly detected lung nodules in 152 patients (1%). Of these, 72 patients were excluded due to lack of follow-up images, and seven were excluded because they did not receive a conclusive clinical diagnosis.

In the final analysis of the remaining 73 patients, the false positive rate was found to be 30.1%. The breakdown showed that 11% had malignancy, 6.9% had active inflammation, 49.3% had postinflammatory sequelae, and 2.7% fell into other categories. This suggested that about 20.6% of incidental lung nodules in groups A, B, and D required further evaluation or treatment. The researchers acknowledged that their study did not provide comprehensive data on the detection and management of lung nodules when using AI-based software. This was partly because clinicians in their hospital had the discretion to consult AI results at their convenience, making it challenging to determine the exact influence of AI on clinical decision-making. Nevertheless, the team plans to further investigate these aspects in future research.

“Our results showed that lung nodules were detected unexpectedly by AI in approximately 1% of initial [chest X-rays], and approximately 70% of these cases were true positive nodules, while 20.5% needed clinical management,” noted lead author Shin Hye Hwang, MD.

Related Links:
Yonsei University College of Medicine
Lunit