Quantitative sonographic feature analysis of clinical infant hypoxia: a pilot study

To determine whether textural features derived from sonographic pixel intensities differ significantly between healthy infants and infants who have had acute clinical hypoxic episodes.

Neurosonographic and calibration phantom-processed image data were evaluated prospectively from 9 infants (age range, 1 to 163 days) with at least 1 episode of hypoxia and compared with image data from a control population of 16 healthy infants (age range, 1 to 191 days). Custom software was used to make 45 textural feature measurements on 40 x 40-pixel regions of interest within brain parenchyma in the distribution of each major cerebral artery, the thalami, and the cerebellum and in a tissue-mimicking calibration phantom. Means comparison testing was followed by logistic regression to assess statistical variation between the patients and the control group.

Nine of 45 textural features showed statistically significant differences between mean values comparing the two groups. Mean gray level was the most sensitive predictor of differences between the two populations (mean gray level for healthy subjects was 46.8; mean gray level for patients was 56.3). An average of mean gray values in areas supplied by the posterior cerebral arteries and the cerebellum was even more sensitive for differentiating healthy subjects from patients.

Quantitative sonographic textural feature analysis showed differences between the brains of healthy infants and those of infants with clinical hypoxia.