Cardiotocography (CTG) is a doppler ultrasound–based mostly approach used throughout being pregnant and labor to observe fetal well-being by recording fetal coronary heart fee (FHR) and uterine contractions (UC). CTG might be finished constantly or intermittently, with leads positioned both externally or internally. Exterior CTG entails using two sensors positioned on the birthing dad or mum’s stomach: an ultrasound transducer positioned above the fetal coronary heart place to observe FHR, and a tocodynamometer (stress sensor) positioned on the fundus of the uterus to measure UC.
At the moment, suppliers interpret CTG recordings utilizing pointers like these from the Nationwide Institute of Baby Well being and Human Improvement (NICHD; pointers) or the Worldwide Federation of Gynecologists and Obstetricians (FIGO; pointers). These requirements outline completely different patterns within the CTG and FHR traces which will point out fetal misery.
Right now we current work from our current paper, ”Improvement and analysis of deep studying fashions for cardiotocography interpretation”, wherein we describe analysis on our new machine studying (ML) mannequin that may present goal interpretation help to well being suppliers to scale back burden and doubtlessly enhance fetal outcomes. Utilizing an open-source CTG dataset, we develop end-to-end neural network-based fashions to foretell measures of fetal well-being, together with each goal (fetal arterial wire blood pH, i.e., fetal acidosis) and subjective (fetal Apgar scores) measures. Given the potential excessive stakes nature of the use-case if utilized in a medical setting, we carry out in depth evaluations to look at how the mannequin performs with various inputs, together with FHR solely, FHR+UC, and FHR+UC+Metadata.