Abstract
BACKGROUND: White matter periventricular echogenicity detected in neonatal cranial ultrasound (US) scans may be followed by cystic periventricular leukomalacia (PVL) or might resolve. Serial US are therefore needed to clarify the outcome. Texture analysis is a computerized method to analyze images and distinguish findings that are undetectable by the human eye. OBJECTIVE: To test whether texture analysis can differentiate echogenicities that resolve from those that develop cystic PVL. MATERIAL AND METHODS: Neonates with echogenicities on their initial US scan were studied; texture analysis was performed on the coronal and sagittal sections. Texture parameters were entered into a linear discriminant analysis (LDA) to classify the scans along two axes called most discriminant features (MDF) 1 and 2. RESULTS: We studied twenty infants with periventricular echogenicities on initial scans; ten of them later resolved (group A), while the other ten infants developed cystic PVL (group B). The classification accuracy was 66% and 82% for group A and B on sagittal sections, and 75%, and 80% on coronal. In the coronal and sagittal plane respectively, a MDF1 value of 0.98 and 0.24 and an MDF2 value of 0.86 and 0.001 provided the best sensitivity, specificity, positive and negative likelihood ratio. CONCLUSION: Texture analysis is a promising objective tool to early identify which cranial echogenicity that will develop into cystic PVL.