Using Fetal Myocardial Velocity Recordings to Evaluate an AI Platform to Predict High-risk Deliveries

Abstract: Diagnosing abnormal fetal cardiac function using ultrasound is a complicated procedure which makes it difficult to obtain high quality results from ultrasound examinations that are performed shortly before delivery. Color tissue Doppler imaging (cTDI) is the echocardiographic technique that has been used to obtain the data for this project. Subtle changes in the fetal cardiac function caused by a variety of complications can possibly be detected using cTDI. Fetuses suffering from these complications are often involved in high-risk deliveries. Combining the data obtained from cTDI with Artificial Intelligence (AI) may improve precision and accuracy when it comes to diagnosing pathological conditions involving fetal cardiac function before delivery. AI uses machines to perform and execute tasks that are characteristic of human intelligence. AI can be achieved by using deep learning. Deep learning uses algorithms called artificial neural networks that are inspired by the biological structure and function of the human brain. The neural networks classify information in a similar manner to the human brain. A platform that uses deep learning can make statements or predictions based on the data fed to it. The AI platform Peltarion uses deep learning to perform tasks. The aim of this project was to use Peltarion to evaluate the possibility of predicting high-risk deliveries with abnormal perinatal outcome by using data obtained by cTDI velocity recordings of the fetal heart. The data included myocardial velocity recordings from 107 pregnancies, out of the 107 pregnancies 82 of the babies were born healthy while 25 babies had an adverse perinatal outcome. The data was uploaded in the platform and three models were built and trained in order to evaluate the performance of the platform using the data. The parameters that have been used to determine the results are loss, accuracy and precision. The results showed that the accuracy parameter was measured to be 0.8 in all cases which means that the model correctly predicts if a fetal heart is healthy or likely to have an adverse outcome 80% of the time. The precision parameter was measured to be around 0.4 which means out of all the times the model predicted a fetal heart to have an adverse outcome, only 40% truly had an adverse outcome. It was concluded that a substantially larger amount of evenly distributed data is required to appropriately evaluate the possibility of using fetal myocardial velocity recordings as data for the AI platform Peltarion to predict high-risk deliveries.