Prediction of periventricular leukomalacia. Part I: Selection of hemodynamic features using logistic regression and decision tree algorithms

摘要: Objective: Periventricular leukomalacia (PVL) is part of a spectrum cerebral white matter injury which associated with adverse neurodevelopmental outcome in preterm infants. While PVL common neonates cardiac disease, both before and after surgery, it less older infants disease. Pre-, intra-, postoperative risk factors for the occurrence are poorly understood. The main objective present work to identify potential hemodynamic complex heart disease using logistic regression analysis decision tree algorithms. Methods: arterial blood gas data (monitoring variables) collected intensive care unit Children's Hospital Philadelphia were used predicting PVL. Three categories datasets 103 used-(1) original without any preprocessing, (2) partial keeping admission, maximum minimum values monitoring variables, (3) extracted dataset statistical features. as inputs forward stepwise select most significant variables predictors. selected features then induction algorithm generating easily interpretable rules prediction Results: sets analyzed SPSS identifying statistically predictors (p<0.05) through their correlations. classification success Case 3 was best sensitivity (SN), specificity (SP) accuracy (AC) 87, 88 87%, respectively. identified features, when algorithms, gave SN, SP AC 90, 97 94% training 73, 58 65% test. mainly included pressure, systolic diastolic, pressures pO"2 pCO"2, like average, variance, skewness (a measure asymmetry) kurtosis abrupt changes). Rules generated automatically Conclusions: proposed approach combines advantages (regression analysis) mining techniques (decision tree) generation prediction. extends an earlier research [Galli KK, Zimmerman RA, Jarvik GP, Wernovsky G, Kuijpers M, Clancy RR, et al. surgery. J Thorac Cardiovasc Surg 2004;127:692-704] form expanding feature set, additional prognostic (namely pCO"2) emphasizing temporal variations addition upper or lower values, rules. further investigated Part II selection computational intelligence.