(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 of the Children's Hospital of Zhejiang University School of Medicine is quietly equipped with an artificial intelligence software. Within a few seconds, a child's left hand radiograph was automatically identified and bone age was read. The maximum gap between machine-reading results and higher-level doctors was within 3 months.
This is the first artificial-intelligent bone-age reader in China that has landed at a hospital. On September 25, the Children's Hospital of Zhejiang University School of Medicine (hereinafter referred to as "Zhejiang Children's Hospital") and Yitu Technology announced their strategic partnership.
With the aid of bone-age data from more than 10,000 healthy children's physical examinations in the Children's Hospital of Zhejiang Province, the use of deep-learning technology by Tuituo Technology has trained the skeletal-age reader.
Shu Qiang, dean of the Children's Hospital of Zhejiang University School of Medicine and deputy secretary of the Party Committee, referred to the cooperation as a "very standardized and formal artificial intelligence project."
According to Fu Junfen, deputy dean of the Children's Hospital of Zhejiang University School of Medicine and director of the project, the Children's Hospital of Zhejiang Province and Yitu Technology cooperated in the form of scientific research projects, and was carried out after the hospital ethics committee and other parties agreed.
Fu Junfen introduced it to the press. At present, this “ruler†for measuring artificial intelligence has been set up. Next, this ruler will also be verified and corrected in the normal crowds of schools and other scenes, and expect to use it. Promote to various hospitals across the country.
Fu Junfen and Niu Hao, president of ETO Medical, also expressed their ambition: “To formulate new standards for interpretation of bone age in Chinaâ€.
Bone Age Determination "Dilemma"
In 2000, Dr. Fu Junfen graduated and began to enter the field of pediatric endocrinology. The idea of ​​changing the skeletal age was planted in her mind at that time.
People have two dimensions of physical age and bone age, and bone age can reflect the degree of human growth and development. It is the most basic tool for pediatric visits and can be used to diagnose and monitor children's endocrine diseases and growth disorders.
But measuring bone age has always faced a dilemma: speed and accuracy is the relationship between fish and paws.
In the first line of clinical diagnosis, the method of determining bone age is the GP map method, which looks "old". On the other hand, the doctor held a hand bone X-ray film, and on the other side, a paper image book published in the 1950s was used to compare naked eyes. This set of samples taken from the Americans decades ago covered all common bony indications during the development of the wrist bones. Trained doctors can quickly give bone age.
But the shortcomings can not be ignored: subjectivity is too strong, clinical accuracy is poor. Fu Junfen said that even if it is the same doctor looking at the same map, the gap between the two before and after may also be 3 months; if different doctors look at the same map, the difference may be up to 2 years old. This is detrimental to clinical diagnosis.
It is not that there is no high accuracy method. For example, TW3 is a method favored by scientific research. In the TW3 method, the first step, the doctor will identify the 20 bones of the hand and rate them one by one, for a total of 8 grades. In the second step, the rating will go through a complex mathematical formula and turn it into a numerical value. Then in the curve table, find the bone age corresponding to the value.
How complicated is this? If only relying on the strength of one doctor, it takes 1 to 2 hours to calculate a bone age tablet; if using computer software, instead of the doctor to complete the second step of the work, it takes 15 minutes to 30 minutes each.
Can it be fast and accurate? This is precisely what artificial intelligence is good at.
"Hello" a ruler
A few months ago, Fujitsu Fukui’s vision began to emerge from the security and financial advancement of medical technology.
Artificial intelligence is used for bone age determination. This is an application scenario that many people in the industry have recently targeted. The imaging doctors are optimistic that the artificial intelligence company wants to intervene, and the manufacturers of hospital image archiving and communication systems have begun to lay out related products.
Who can win, in a sense determined by who obtains quality data. “For medical artificial intelligence and even the entire field of artificial intelligence, high-quality labeling data is the core.†Lin Qiang, who is responsible for the IBG project, said: “You have to feed the data. It is useless to throw data, and it requires a lot of labeled data. The marked data is very difficult to obtain."
Ni Hao said that what many people may not have thought is that this is the biggest cost of the artificial intelligence bone age reader. "Hello" is particularly important for the early labeling data of the algorithm. "You can't read the labels until they are accurately measured."
"Labeling is impossible for artificial intelligence companies to do. Requires radiology and endocrinology doctors to repeatedly track, labeling can not be posted wrong." Fu Junfen said, Zhejiang Provincial Children's Hospital Radiology and Endocrinology doctors are almost always put into the label In addition, thousands of manually labeled high-quality bone-age tablets are provided.
Because of the privacy involved, a total of more than 10,000 bone-aging tablets of healthy children were desensitized by the Zhejiang Provincial Children's Hospital Information Department.
On this basis, the machine self-learns, iterates, trains the model, and then feedbacks, artificially scores, tells the machine whether it is “soft†or “oldâ€, and the machine corrects itself.
In the end, this was the birth of a “barometer†that was “fed†based on data from healthy bone age tablets. At the speed of seconds, the machine completes all the steps in the TW3 method, automatically finds the bone defects in the X-ray film, ranks them, and then substitutes them into the formula to compare the bone ages with numerical values.
When applied to the actual clinic, the final judgement of bone age will still be checked by a doctor. From the fastest 15 minutes to several seconds, Fu Junfen delighted doctors can save a lot of time. "China lacks 200,000 pediatricians," she said.
American Motors Replacement,Oil Filters for American Cars,Oil Filters for American Auto Motors,American Motors Replacement Oil Filters
Zhoushan Shenying Filter Manufacture Co., Ltd. , https://www.renkenfilter.com