Automated Deep Learning Detection Algorithms for Fetal Orientation and Placenta Previa

Identifying the correct mode of fetal delivery is critical for ensuring the survival of both the mother and fetus, and it is influenced by fetal orientation and the presence of Placenta Previa (PP). To automate this process, we developed two deep-learning algorithms using Convolutional Neural Networks (CNNs) to classify fetal orientation and identify PP from two-dimensional (2D) Magnetic Resonance Imaging (MRI) slices. Our fetal orientation classifier, Fet-Net, achieved an average classification accuracy of 97.68% on 6120 MRI slices during a 5-fold cross-validation experiment. Our PP classifier, Previa-Net, performed with an average classification accuracy of 96.95% on 420 MRI slices across five random seeds. Both models outperformed state-of-the-art architectures such as VGG, ResNet, and Inception. By combining these two models, we can expedite the fetal exam reading for radiologists and determine the likelihood of surgical delivery based on fetal and placental positions, improving obstetric health care.