Smart Hip Ultrasound Technology in Screening for Infant Developmental Dysplasia of the Hip
10.13471/j.cnki.j.sun.yat-sen.univ（med.sci）.2023.0521
- VernacularTitle:智能髋关节超声技术在婴儿发育性髋关节发育不良筛查中的应用
- Author:
Ying XU
1
;
Hong-kui YU
1
;
Xiao-ying LIN
1
;
Yang ZHAO
1
;
Xing-yi YANG
1
;
Zi-yin HUANG
2
;
Xiao XU
2
Author Information
1. Department of Ultrasound， Shenzhen Bao'an District Maternal and Child Health Hospital， Shenzhen 518133， China
2. Shenzhen Kaili Biomedical Technology Co， Shenzhen 518057， China
- Publication Type:Journal Article
- Keywords:
infant;
developmental dysplasia of the hip （DDH）;
ultrasonography;
deep learning;
automated measurement;
screening
- From:
Journal of Sun Yat-sen University(Medical Sciences)
2023;44(5):870-877
- CountryChina
- Language:Chinese
-
Abstract:
ObjectiveTo develop a deep learning system for early ultrasound screening of developmental dysplasia of the hip （DDH）， a new smart-hip ultrasound technique （S-hip）， and to validate its clinical application. MethodsWe selected 11，100 annotated and reviewed coronal ultrasound images of infant hips between November 2021 and August 2022， 8，100 of which were used for the training set and 3，000 for the test set， to build a S-hip deep learning system. To verify the consistency between the automated measurement by S-hip and the manual measurements by sonographers， 174 standard coronal ultrasound images of 87 infants' bilateral hips were acquired， then α angle， β angle and femoral head coverage （FHC） were measured by S-hip， an ultrasound expert and a resident. The measurement data and the time required for the measurements were recorded and statistically analyzed. Another 100 standard coronal ultrasound images of the hips were randomly selected and measured twice respectively by the ultrasound expert and resident to assess the intra-sonographer repeatability. ResultsThe intraclass correlation coefficient （ICC） （95% CI） values of α angle， β angle and FHC results measured by S-hip and ultrasound expert were 0.799 （0.738， 0.847）， 0.798 （0.737， 0.846） and 0.934 （0.954， 0.975）， respectively. Those values measured by the ultrasound expert and resident were 0.725 （0.645， 0.789）， 0.674 （0.583， 0.748） and 0.931 （0.908， 0.949）， respectively. The mean absolute errors （MAE） of α angle， β angle and FHC results between measurements by S-hip and ultrasound expert were 2.69 °， 4.43 ° and 2.47%， respectively. The time required for measurements by S-hip， ultrasound expert and resident was （1.59±0.36） s， （18.76±2.23） s and （19.45±2.76） s， respectively. The automated measurement by S-hip cost much shorter time than the manual measurements by sonographers and the difference was statistically significant （P<0.001）. The ICC （95% CI） values of α angle， β angle and FHC results between two measurements by the ultrasound expert were 0.943 （0.916， 0.961）， 0.959 （0.940， 0.972）， and 0.981 （0.971， 0.987）， respectively. Those values by the ultrasound resident were 0.884 （0.833， 0.921）， 0.921 （0.884， 0.946）， and 0.962 （0.944， 0.974）. ConclusionThe S-hip based on a deep learning system is a highly reliable automated technique to accurately measure α angle， β angle and FHC. Compared with ultrasound residents， S-hip allows for a more simplified and significantly quicker measurement， which may enhance the widespread use of hip ultrasound screening in infants.
