R/read_muse_xml_ecg.R

In rickeycarter/easyRecg: Simple tools to facilitate reading and writing digital ECG files

#' Read a single XML representation of an ECG exported from MUSE. Returns a matrix with data for 12-leads
#'
#' @param file  Filename for the xml file that will be read
#' @param file_out Filename for exported data.  If NA, no file is written.  Only works if numpyformat is FALSE.
#' @param numpyformat Logical to determine if the array should be returned as a 5000x12 (FALSE) or a 1x5000 x 12 x 1 (TRUE, Default) format.  The latter is the standard input for AI-ECG algorithms
#' @export
#' @examples
#' \dontrun{
#' ecg_1 <- read_muse_xml_ecg("1_MUSSE_ecg.xml")
#' }

read_muse_xml_ecg <- function(file, file_out = NA, numpyformat=TRUE) {
  # Read XML 
  doc <- xml2::read_xml(file)
  
  # Rhythm is the second Wavefrom (Median first)
  rhythm <- xml2::xml_contents(xml2::xml_child(doc, "Waveform[2]"))
  
  # Index of lead data 
  rhythm_names <- xml2::xml_name(rhythm)
  leads <- which(rhythm_names == "LeadData")
  
  # Get sample count.  All leads should be the same.
  first <- xml2::xml_child(rhythm, "LeadSampleCountTotal")[leads[1]]
  samples <- xml2::xml_integer(first)
  
  # Matrix to hold results
  lead_names <- c("I", "II", "III", "aVR", "aVL", "aVF", "V1", "V2", "V3", "V4", "V5", "V6")
  lead_data <- matrix(nrow = samples, ncol = length(lead_names))
  colnames(lead_data) <- lead_names
  
  # Loop through each leads and extract data
  for (l in leads) {
    lead <- xml2::as_list(rhythm[l][[1]])
    id <- lead$LeadID[[1]]
    amp_per_byte <- as.numeric(lead$LeadAmplitudeUnitsPerBit[[1]])
    waveform <- lead$WaveFormData[[1]]
    bin <- base64enc::base64decode(waveform)
    data <- readBin(bin, integer(), samples, size = 2) * amp_per_byte
    lead_data[,id] <- data
  }
  
  # III = II - I
  lead_data[,"III"] <- lead_data[,"II"] - lead_data[,"I"]
  
  # aVR = -(I + II)/2
  lead_data[,"aVR"] <- -(lead_data[,"I"] + lead_data[,"II"])/2
  
  # aVL = I - II/2
  lead_data[,"aVL"] <- lead_data[,"I"] - lead_data[,"II"]/2
  
  # aVF = II - I/2
  lead_data[,"aVF"] <- lead_data[,"II"] - lead_data[,"I"]/2
  
  # Do we really want integer? Matches python script.  Would round() be better? Or no rounding? 
  lead_data <- apply(lead_data,2,as.integer)
  
  # export a file if the 5000x12 array is the target output.  If numpyformat (4 dimensional array), this is skipped.
  if (!is.na(file_out) & !numpyformat) {
    utils::write.table(lead_data, file = file_out, row.names = FALSE)
  }
  
  ## Add flag to reshape the array to match standard ECG AI inputs
  if (numpyformat){
    dim(lead_data) <- c(1, 5000, 12, 1)
  } 
  lead_data
}