제목: Deep learning aids screening of ileocolic intussusception on radiographs of young children
For pediatric radiology, deep learning-based algorithms were first developed and applied in bone age assessment. However, further adaptation of deep learning in pediatric radiology was stagnated by significant potential limitations, such as requirement of bid data set, necessity of accurate labeling, and different data according to the normal aging process of children, as mentioned by Michael M. Moore et al. in Pediatric Radiology. Even with those limitations, Hyun Joo Shin, MD, PhD, clinical assistant professor at Yonsei University College of Medicine, and pediatric radiologist in the Department of Radiology in the Yonsei University Health System in Seoul, Korea, said that “Deep learning-based algorithm can aid screening of intussusception using abdominal radiography in young children”. Dr. Shin and Dr. Sungwon Kim developed and first adopted deep learning based algorithm for the pediatric abdominal radiographs to detect intussusception.
Development and diagnostic performance of deep learning based algorithm for detection of pediatric intussusception
They developed deep learning based algorithm using training set from total 681 children (≤5 years old, 242 children with intussusception) who underwent abdominal radiograph and ultrasonography for suspicion of intussusception from March 2005 to December 2017. A YOLOv3-based algorithm was developed to recognize the rectangular area of the right abdomen and to diagnose intussusception to overcome relatively small number of data set in pediatric radiographs. From the time independent validation set composed of 75 children (≤5 years old, 25 children with intussusception, from January to August 2018), the sensitivity of the algorithm was higher compared with that of the radiologists (0.76 vs. 0.46, p = 0.013), while specificity was not different between the algorithm and the radiologists (0.96 vs. 0.92, p = 0.32).
Clinical implication of deep learning based approach for pediatric intussusception
Dr. Shin said thatthe algorithm might be used as screening tool for selecting children who need additional US examinations or who need to be referred to tertiary hospitals and this could be done with low false-negative rates. This could be a method for reducing unnecessary emergent US examinations and lowering the burden on radiologists and also patients. Because this study first used deep learning for pediatric abdominal radiographs and emergency disease of children, further external and clinical validation is expected to be processed afterward.
For More Information,
Access the study “Performance of deep learning-based algorithm for detection of ileocolic intussusception on abdominal radiographs of young children”[S1] , and read the accompanying minisymposium “Machine learning concepts, concerns and opportunities for a pediatric radiologist”[S2]
참고 url: https://www.nature.com/articles/s41598-019-55536-6
[S2]Moore MM, Slonimsky E, Long AD, Sze RW, Iyer RS. Machine learning concepts, concerns and opportunities for a pediatric radiologist. Pediatr Radiol 2019; 49: 509-516 [PMID:30923883 DOI:10.1007/s00247-018-4277-7]