RESUMO
En este trabajo se presenta un nuevo conjunto de indicadores de severidad que combinan diversos rasgos craneales para cuantificar las craneosinostosis aisladas de tipo sagital y metópica. La utilidad de los indicadores se evaluó examinando las tomografías computarizadas del cráneo de un grupo de infantes afectados por craneosinostosis aislada y un grupo de infantes no afectados. La base de datos contiene estudios de 90 pacientes con craneosinostosis sagital, 40 con craneosinostosis metópica y 60 pacientes no afectados. Los indicadores de severidad se obtienen a partir de un conjunto de indices de severidad por medio de un método estadístico de regresión logística regularizada conocido como red elástica. Los índices de severidad son medidas univariadas de forma que se calculan a partir de tres planos de análisis. Los planos se estiman a partir de referencias anatómicas cerebrales radiológicamente identificables. El desempeño de los indicadores se midió estimando el grado de separación lineal (GSL), que cuantifica la capacidad de un indicador para distinguir cráneos sagitales o metópicos de cráneos no afectados. Los indicadores de severidad propuestos alcanzan un GSL del 95.83% y 98.9% en las poblaciones sagitales vs. controles y metópicos vs. controles, respectivamente. Los resultados obtenidos en este trabajo sugieren que es posible construir indicadores multivariables de severidad que son clínicamente reproducibles y cuantifican efectivamente aspectos de la morfología craneal codificada por medio de un conjunto de índices de severidad.
This work develops a new set of severity scores that combine several cranial features in order to quantify sagittal and metopic craniosynostosis. Computed tomography head scans were obtained from 90 children affected with single-suture sagittal synostosis, 40 children with single-suture metopic synostosis, and 60 age-matched nonsynostotic controls. Tridimensional reconstructions of the skull were used to trace image analysis planes defined in terms of skull-base plane and internal landmarks. For each patient, a new set of descriptive measures or severity indices of skull shape malformation were computed. A statistical classification approach (regularized logistic regression) was used for combining individual severity indices into summarizing severity scores. The linear separation index that measures the ability of a classification function to separate the affected (sagittal or metopic) and nonsynostotic populations was used to evaluate the severity scores. The proposed scores are sensitive measures of the calvarial malformation that achieve linear separation indices of 95.83% and 98.9% for sagittal vs. control and metopic vs. control populations, respectively. As opposed to individual severity indices, the summarizing severity scores encapsulate a number of distinctive calvarial features associated with sagittal and metopic synostoses crania. The proposed scores enable quantitative analysis in clinical settings of skull features observed in isolated sagittal and metopic synostoses that may not be accurately detected by separate analysis of individual severity indices.
RESUMO
This study presents evidence suggesting that electrophysiological responses to language-related auditory stimuli recorded at 46weeks postconceptional age (PCA) are associated with language development, particularly in infants with periventricular leukomalacia (PVL). In order to investigate this hypothesis, electrophysiological responses to a set of auditory stimuli consisting of series of syllables and tones were recorded from a population of infants with PVL at 46weeks PCA. A communicative development inventory (i.e., parent report) was applied to this population during a follow-up study performed at 14months of age. The results of this later test were analyzed with a statistical clustering procedure, which resulted in two well-defined groups identified as the high-score (HS) and low-score (LS) groups. The event-induced power of the EEG data recorded at 46weeks PCA was analyzed using a dimensionality reduction approach, resulting in a new set of descriptive variables. The LS and HS groups formed well-separated clusters in the space spanned by these descriptive variables, which can therefore be used to predict whether a new subject will belong to either of these groups. A predictive classification rate of 80% was obtained by using a linear classifier that was trained with a leave-one-out cross-validation technique.