On April 19th, the first Human Machine Reading Competition (hereinafter referred to as “Human Machine Competitionâ€) with the theme of “The Nature of Ultrasound Image of Thyroid Nodules†was held at Beijing Friendship Hospital affiliated to Capital Medical University. This is an innovative attempt by artificial intelligence in the medical field, and it also indicates that “AI+Medical†is gradually approaching.
The human players who participated in the "Human-Machine Film Competition" were professional doctors from 200 imaging departments from dozens of top three hospitals such as Friendship Hospital, Union Hospital, Beijing Medical Third Hospital, and Beijing Cancer Hospital. The representative of artificial intelligence was played by The “Diagnosis-medical Image-Assisted Diagnostic Tool†(hereinafter referred to as “visual diagnosisâ€) developed by the Beijing University Computing Center “Qihongtu†artificial intelligence R&D team.
According to the relevant person in charge, “Diagnosis†is based on deep learning and image processing algorithms. In the data, we have learned the rich and accurate thyroid nodule ultrasound image data of Beijing Friendship Hospital, and got a team of professional doctors in research and development and clinical application. Guidance. At present, the accuracy rate of the “diagnosis and diagnosis†in the B-ultrasound image internal diagnosis is 76%, which is comparable to the diagnostic level of the 5-year experience attending physician.
"Human Machine Competition" scene
The exam questions for this competition are 100 sets of ultrasound images of thyroid nodules that have been provided by Beijing Friendship Hospital for the determination of pathological results ("gold standard"). Both the “visual diagnosis†and the human doctor’s players must make a benign and malignant judgment on the 100 sets of ultrasound images. Each time the answer is 1 point, all the answers are 100 points. The conference team scored the same scoring standard, and finally ranked the competition according to the correct rate, and set two awards, individual award and group award.
During the competition, “visual diagnosis†showed excellent stability. Among the 85 contestants, the “visual diagnosis†accuracy rate was 73%, ranking sixth (the accuracy rate of the first place was 76%). According to the team ranking, “visual diagnosis†defeated all participating hospital teams with a slight advantage of accuracy exceeding 0.3%. What is more worth mentioning is that the “visual diagnosis†answered 100 questions in only 514 seconds, while Zhang Mingbo, the attending physician of the 301 hospital ultrasound diagnosis department with the first 8 years of “reading film†experience, took 909 seconds.
Tubular Cam Lock,Cam Locks,Cabinet Locks,Cabinet Latches
Ningbo Hengchieh Locking Technology Co., Ltd. , https://www.hengchieh.com