Ensemble Learning Using Individual Neonatal Data for Seizure Detection.

OBJECTIVE: Sharing medical data between institutions is difficult in practice due to data protection laws and official procedures within institutions. Therefore, most existing algorithms are trained on relatively small electroencephalogram (EEG) data sets which is likely to be detrimental to prediction accuracy. In this work, we simulate a case when the data can not be shared by splitting the publicly available data set into disjoint sets representing data in individual institutions. METHODS AND PROCEDURES: We propose to train a (local) detector in each institution and aggregate their individual predictions into one final prediction. Four aggregation schemes are compared, namely, the majority vote, the mean, the weighted mean and the Dawid-Skene method. The method was validated on an independent data set using only a subset of EEG channels. RESULTS: The ensemble reaches accuracy comparable to a single detector trained on all the data when sufficient amount of data is available in each institution. CONCLUSION: The weighted mean aggregation scheme showed best performance, it was only marginally outperformed by the Dawid-Skene method when local detectors approach performance of a single detector trained on all available data. CLINICAL IMPACT: Ensemble learning allows training of reliable algorithms for neonatal EEG analysis without a need to share the potentially sensitive EEG data between institutions.

Assessing Velocity and Directionality of Uterine Electrical Activity for Preterm Birth Prediction Using EHG Surface Records.

The aim of the present study was to assess the capability of conduction velocity amplitudes and directions of propagation of electrohysterogram (EHG) waves to better distinguish between preterm and term EHG surface records. Using short-time cross-correlation between pairs of bipolar EHG signals (upper and lower, left and right), the conduction velocities and their directions were estimated using preterm and term EHG records of the publicly available Term-Preterm EHG DataSet with Tocogram (TPEHGT DS) and for different frequency bands below and above 1.0 Hz, where contractions and the influence of the maternal heart rate on the uterus, respectively, are expected. No significant or preferred continuous direction of propagation was found in any of the non-contraction (dummy) or contraction intervals; however, on average, a significantly lower percentage of velocity vectors was found in the vertical direction, and significantly higher in the horizontal direction, for preterm dummy intervals above 1.0 Hz. The newly defined features-the percentages of velocities in the vertical and horizontal directions, in combination with the sample entropy of the EHG signal recorded in the vertical direction, obtained from dummy intervals above 1.0 Hz-showed the highest classification accuracy of 86.8% (AUC=90.3%) in distinguishing between preterm and term EHG records of the TPEHGT DS.