PT - JOURNAL ARTICLE AU - Nagy, Z AU - Lagercrantz, H AU - Forssberg, H AU - Chu, C TI - PO-0427 Investigating The Use Of Support Vector Machine Classification On Structural Brain Images Of Preterm–born Teenagers As A Biological Marker AID - 10.1136/archdischild-2014-307384.1069 DP - 2014 Oct 01 TA - Archives of Disease in Childhood PG - A385--A385 VI - 99 IP - Suppl 2 4099 - http://adc.bmj.com/content/99/Suppl_2/A385.1.short 4100 - http://adc.bmj.com/content/99/Suppl_2/A385.1.full SO - Arch Dis Child2014 Oct 01; 99 AB - Background and aims Preterm birth is identified as a risk factor for brain development. We investigate the utility of support vector machine classification as a biological marker for outcome after preterm birth. Methods We trained a linear support vector machine using the grey matter segment (Figure 2) of a 3D MR image (resolution 0.98 × 0.98 × 1.5 mm3) collected from 143 individuals (69 controls) at adolescence. Subsequently, each individual was automatically classified preterm/control. Using birth weight, gestational age or IQ as independent variables and the prediction score (i.e. distance to the decision boundary) as dependent variable we quantified correlations. Abstract PO-0427 Figure 2 Results Correct classifications occurred 93% of the time. The correlation with the prediction score was stronger for birth weight (R = –0.51, p < 0.000001) than gestational age (R = –0.24, p < 0.04) but wasn’t significant within the control group only. IQ was significantly correlated with the prediction score (R = –0.30, p < 0.001). Fig1 depicts the prediction scores for both groups (Top). For the subset for which it was available the IQ scores were used to colour code the scatter plot (bottom). Abstract PO-0427 Figure 1 Conclusions The 93% correct classification is comparable to studies involving individuals with e.g. Alzheimers. The current study is a proof-of-principle, testing the necessary condition whether SVM classification could identify individuals who were born preterm based on a single MR image. The long-term goal of this method is predicting outcome by classifying preterm individuals as having a more “control-like” or “preterm-like” brain. Such information could be used to predict neurological/psychological scores and outcome.