Early Prediction of Periventricular Leukomalacia Using Quantitative Texture Analysis of Serial Cranial Ultrasound Scans in Very Preterm Infants

We compared texture parameters of serial cranial ultrasound (cUS) images of periventricular leukomalacia (PVL) and normal periventricular echogenicity (PVE) in very preterm infants and evaluated the early predictive values of texture analysis (TA) for PVL. Ten individuals with PVL and 10 control individuals with PVE assessed with an initial cUS within 1 wk of birth and follow-up cUS at 2–3 and 4–6 wk of life were included. TA was performed on the region of interest of PVE at the parieto-occipital area on serial cUS. Opposite changes in variance were obtained between the first two cUS sessions in both groups (p = 0.017 in PVL and p = 0.005 in PVE). The variance-to-mean ratio (VMR) between the second and first cUS sessions differed (p = 0.016) and reliably stratified the groups (area under the receiver operating characteristic curve: 0.820, 95% confidence interval: 0.587–1.000, sensitivity: 100%, specificity: 60%). TA of serial cUS helps to predict PVL within 3 wk of life.