(Original title: From the quarter hour to several seconds, China's first artificial intelligence bone age reader was put into use)
News reporter Wang Yingying
In the past two weeks, the Department of Radiology at the Children’s Hospital of Zhejiang University School of Medicine has quietly integrated an artificial intelligence software. Within just a few seconds, the system can automatically analyze a child’s hand X-ray and determine their bone age. The difference between the AI's readings and those of senior doctors is no more than three months.
This marks the debut of China’s first AI-based bone age reader in a hospital setting. On September 25, the Children’s Hospital of Zhejiang University School of Medicine (Zhejiang Children’s Hospital) and Yitu Technology announced a strategic partnership.
Using bone age data from over 10,000 healthy children examined at the hospital, Yitu Technology applied deep learning technology to train its skeletal age detection system.
Shu Qiang, dean of the Children’s Hospital of Zhejiang University School of Medicine and deputy secretary of the Party Committee, called this collaboration a “very standardized and formal AI project.â€
According to Fu Junfen, deputy dean of the hospital and director of the project, the cooperation was conducted through a scientific research initiative that received approval from the hospital’s ethics committee and other relevant parties.
Fu Junfen explained to the press that the AI tool for measuring bone age has now been set up and will be further tested and refined in real-world settings such as schools. The goal is to eventually roll it out across hospitals nationwide.
Fu Junfen and Niu Hao, president of ETO Medical, expressed their ambition: “To establish new standards for interpreting bone age in China.â€
The Bone Age Dilemma
In 2000, Dr. Fu Junfen began her career in pediatric endocrinology, planting the idea of improving bone age assessment in her mind.
Humans have two dimensions of age: chronological age and bone age. Bone age reflects the level of growth and development, serving as the most basic tool in pediatric care. It helps diagnose and monitor endocrine diseases and growth disorders in children.
However, determining bone age has long faced a dilemma—speed versus accuracy.
Traditionally, doctors use the GP method, which involves comparing a child’s hand X-ray with a paper reference book published in the 1950s. While experienced doctors can quickly estimate bone age, the process is highly subjective and often inaccurate.
Fu Junfen noted that even the same doctor might give a result differing by three months when reviewing the same image on different occasions. When different doctors evaluate the same X-ray, the discrepancy could be as high as two years, which is problematic for clinical decisions.
Although there are more accurate methods like the TW3 approach, they are time-consuming. Doctors must assess 20 bones in the hand and assign each a grade, then plug the numbers into complex formulas before matching them to a curve table. This process can take up to two hours manually, or 15–30 minutes with software assistance.
Can speed and accuracy coexist? Artificial intelligence may hold the key.
“Hello†a Ruler
A few months ago, Fujitsu Fukui started exploring medical applications of AI beyond security and finance.
Using AI for bone age determination has become a hot topic in the industry. Imaging specialists are optimistic about AI involvement, and manufacturers of hospital imaging systems are already developing related products.
As Lin Qiang, who leads the IBG project, said, “High-quality labeled data is the core of medical AI and even general AI. You need to feed the model with good data. Just throwing data in won’t work. Labeled data is hard to get.â€
Niu Hao added that many people overlook the cost of building an AI-based bone age reader. “The early labeling data is crucial. You can't label without accurate measurements,†he said.
“Labeling is not something AI companies can do alone. It requires radiologists and endocrinologists to repeatedly track and label data carefully,†Fu Junfen said. Zhejiang Provincial Children’s Hospital doctors have contributed thousands of manually labeled, high-quality bone age images.
To protect privacy, the hospital’s Information Department desensitized over 10,000 bone age X-rays from healthy children.
Based on this dataset, the machine learns, iterates, trains the model, and receives feedback from human experts to correct its results.
Eventually, this led to the creation of a “ruler†trained on data from healthy children. It can complete all steps of the TW3 method in seconds, automatically detecting bone features, ranking them, and calculating the bone age.
In practice, the final decision still rests with the doctor. From 15 minutes down to just a few seconds, Fu Junfen says this AI tool can save significant time for medical professionals. “China lacks 200,000 pediatricians,†she said.
Shenzhen ChengRong Technology Co.,Ltd. , https://www.laptopstandsupplier.com