Towards automation of transcranial Doppler ultrasound data in pediatric cerebral malaria

Abstract

ObjectiveTo work towards automating Transcranial Doppler (TCD) imaging data in cerebral malaria (CM) patients. MethodsFunctional principal component analysis (FPCA) was used to identify important features of TCD data and quantify its variation explained by each feature. Cox regression was used to test the association between TCD measurements and mortality hazards rate. A functional clustering algorithm was also employed to cluster TCD waveform data into distinct phenotypic groups. Fisher's exact test and one-way analysis of variance (ANOVA) were used to test the association between the TCD-defined phenotypic groups and 14 variables including demographic data, laboratory measurements, and clinical features. ResultsThe systolic, diastolic, and mean velocities as well as the Windkessel notch together explained more than 98% of total TCD data variation in the middle cerebral arteries (MCAs) and the basilar artery (BA). TCD measurements were associated with the survival hazards rates after adjusting for age and gender (chi-squared test statistic = 31.56, p-value = 0.025). A functional data clustering algorithm identified 7 clusters for the MCA TCD data. No associations between phenotypic groups and demographics, clinical features, or laboratory measurements were detected. ConclusionTCD may be useful as a diagnostic and monitoring tool in CM. Developing an algorithm-driven approach to interpreting the waveforms will enhance the scalability of this tool.