Table 3.
Año | Studied population | Method | Automatic segmentation | Accuracy, reliability and/or gold standard comparation | |
---|---|---|---|---|---|
Csutak et al. | 2003 | 250 healthy infants. Age range from 26 weeks of GA to 6 months | Traced the area of seven to ten sections for each ventricle and multiply by the distance between them | No | No |
Kishimoto et al. | 2016 | 26 preterm infants. 11 with moderate to severe IVH | Motorized device. 2D US images used to build a 3D US. Images were manually segmented in parallel sagittal slices 1 mm apart | No | Yes. Compared 3D US to 3D MRI manual segmentation |
Qiu et al. | 2017 | 14 preterm infants. 4 with moderate to severe IVH. 70 3D US | Motorized device (same as in Kishimoto et al.). 3D US built from 2D US. Developed a multiatlas segmentation method | Yes | Yes. Mean DSC of 86.4% ± 8.3% |
Gontard et al. | 2020 | 10 preterm infants with PHVD. 152 3D US scans | Specific 3d transducer (SVNA5-8B, 5–8 MHz). Developed a deep learning automatic segmentation method | Yes | Yes. Mean DSC 87% in training data. Mean DSC 80% in validation data |
GA gestational age, MRI magnetic resonance imaging, DSC dice similarity coefficient, PHVD post haemorrhagic ventricular dilatation.