AI Developed by Training on Over 200,000 Pairs of Follow-up X-ray Images "80% Accuracy"

Professor Junbeom Seo from the Department of Radiology and Professor Namguk Kim from the Department of Convergence Medicine at Seoul Asan Medical Center announced on the 31st that they have developed an artificial intelligence model that diagnoses changes in diseases using over 200,000 pairs of chest X-ray images taken for follow-up examinations.

Professor Junbeom Seo of Radiology at Asan Medical Center, Seoul (left), Professor Namguk Kim of Convergence Medicine

Chest X-ray examination is a basic test performed to diagnose lung or heart diseases and is used as a follow-up test to confirm changes in disease or treatment effects. Until now, research on artificial intelligence diagnosing diseases from a single X-ray image has been actively conducted, but there was no AI capable of diagnosing changes from a pair of X-ray images taken over time.

Professors Junbeom Seo and Namguk Kim’s team developed AI that diagnoses changes in diseases using 203,056 pairs of chest X-ray images from patients who underwent chest X-ray examinations between 2011 and 2018. The research team analyzed the reading process of radiology specialists when interpreting chest X-ray images and trained the AI to mimic this process.

Furthermore, the team introduced an anatomical structure alignment module to enable the AI to focus on similar regions between past and current X-ray images during interpretation. They also employed a multi-task learning technique to allow the AI to understand diseases and evaluate changes in disease conditions.

The research team then evaluated the AI’s accuracy through internal validation using 1,620 pairs of X-ray images and external validation using datasets of 215 and 267 pairs. As a result, the prediction accuracy was approximately 80% in both internal and external validations, comparable to the accuracy of second- and third-year radiology residents.

Professor Junbeom Seo of the Department of Radiology at Seoul Asan Medical Center stated, “Previous AI research mainly focused on diagnostic assistance technology that detects diseases from a single X-ray image, but this study can detect changes in diseases during follow-up examinations, so it is expected to be applicable in actual clinical settings in the future.”

Professor Namguk Kim of the Department of Convergence Medicine at Seoul Asan Medical Center said, “This study is significant not only because the AI mimicking the reading process of radiology specialists can interpret changes in diseases, but also because it improved diagnostic accuracy through a large-scale dataset of over 200,000 pairs.”

Meanwhile, the results of this study were recently published in ‘Medical Image Analysis,’ one of the prestigious journals in the field of medical imaging.

This content was produced with the assistance of AI translation services.