This is a problem ordered by Hanoi Medical University Hospital, and was carried out by Bui Van Son (born in 2000, from Thanh Hoa), a new engineer at Hanoi University of Science and Technology, within 3 years, under the guidance of Dr. Nguyen Hong Quang, Head of the Biomedical Informatics Research Department, International Research Center for Artificial Intelligence (BK.AI).
Bui Van Son is a student of Computer Engineering. Because of his passion for research, from his second year of university, Son applied to join the Biomedical Informatics lab of Dr. Nguyen Hong Quang. Here, the male student had the opportunity to learn about some problems applying AI in the medical field such as decoding genes, finding drug responses, image processing, etc.
While most of his classmates tend to work on websites and apps, focusing on artificial intelligence in the biomedical field, according to Son, also has many difficulties.
“In addition to having to research and participate in artificial intelligence courses, I also had to relearn knowledge related to gene sequencing, DNA, mRNA, PCR testing process... to serve the research.”
During his first time in the lab, under the guidance of his teacher, Son participated in finding new drugs to respond to, prevent and inhibit cancer cells. In addition, the male student also did problems on image processing and identifying faulty panels on solar panels.
In mid-2022, Son was proposed by his teacher to participate in a project ordered by Hanoi Medical University Hospital, related to "Determining the fetal nuchal translucency using ultrasound images".
To fully understand the technical terms, Son had to spend 2 months talking with doctors at the hospital, watching ultrasound videos of fetuses to understand the nuchal translucency measurement process. Gradually, Son understood the meaning, purpose of the topic and how the research results could be applied to real life.
According to Son, determining the width of the nuchal translucency can help doctors make early diagnoses of fetal abnormalities before birth. However, ultrasound measurement of nuchal translucency is currently still performed manually. This technique depends on the doctor's experience, so there are many potential errors.
“Sometimes, just a 0.1-0.2mm error can lead to different diagnostic results, which affects the counseling process for pregnant women,” Son said.
Therefore, the male student spent time studying a dataset of about 1,200 fetal ultrasound images that had been labeled with the nuchal translucency region by leading doctors at Hanoi Medical University Hospital, and at the same time built and developed deep learning models and image processing algorithms suitable for the problem. However, in the beginning, the results were not very promising.
“There was a period of 2-3 weeks, I tried many advanced image processing models but still could not improve the results. There have been studies in the world using artificial intelligence to measure the nape of the neck, but there has not been any study that clearly shows the nape of the neck and its width in mm,” Son said.
Every time they couldn’t find a new direction, the teacher and student had to sit down together to “untangle” and improve each step of the algorithm. According to Dr. Nguyen Hong Quang, Son is a persistent and hard-working person. In order to solve that problem, Son actually had to do nearly 20 different sub-problems.
“The more Son works, the more he finds problems that need to be solved and actively implements them, thanks to which the results improve day by day,” said Dr. Quang.
After nearly 2 years of research, Bui Van Son has proposed a new method to determine and measure the fetal nuchal translucency using 2D ultrasound images with an error of 0.4mm compared to the manual measurement method of the doctor. This measurement method helps reduce the cost and effort of doctors in performing the measurement and serves as a basis for doctors to be able to check their ultrasound measurement process.
Son's research results were also tested at Hanoi Medical University Hospital and were evaluated as feasible. Son then brought these results to the Student Scientific Research Competition of Hanoi University of Science and Technology and won second prize.
According to Son, although this was a new study, the accuracy of the method at that time was not absolute.
Therefore, after the competition, Son continued to develop and improve the algorithm. The male student built a website and application right on the phone. Just upload the image to the system, the model will accurately measure data on nuchal translucency, the safety threshold in just 5-7 seconds. This method can also handle cases that are difficult to identify in blurry, unclear ultrasound images, minimizing unnecessary errors during the ultrasound measurement process.
Seeing the positive results, Bui Van Son and his student have completed a scientific research paper to submit to the international journal of Biomedical Informatics. However, Son also affirmed that this model does not play the role of replacing doctors but is only a supporting tool for doctors to have a basis to determine the brightness more accurately.
Starting to work with Son in his second year, Dr. Nguyen Hong Quang assessed Son as someone who is passionate about scientific research. The first tasks assigned to him at the lab were application programming and web programming, which Son did very quickly and efficiently. After that, because of his passion for AI, Son researched and applied and solved the assigned problems very well.
“With the study “Determining the fetal nuchal translucency using ultrasound images”, the current accuracy of the model is equivalent to that of the leading ultrasound doctors in Vietnam today. However, to put it into practical application depends a lot on factors related to the hospital's operating procedures and the medical examination and treatment procedures of the Ministry of Health”, Dr. Nguyen Hong Quang assessed.
Source
Comment (0)