Fetal MRI has the potential to complement US imaging and improve fetal development assessment by providing more accurate volumetric information about the fetal structures. However, volumetric measurements require manual delineation of the fetal structures, which is time consuming, annotator-dependent, and error-prone. State-of-the-art automatic segmentation methods for volumetric scans are based on deep neural networks and require a large, high-quality dataset of expert-validated annotations, which is very difficult to obtain.
In this talk, we will explore the usefulness of several tools for creating cost-effective segmentation models using small, annotated datasets and additional unlabeled data. First, we will see the effectiveness of Self-training or Pseudo-labeling method for segmentation sequence transfer. Then, we will see that the use of partial annotations compared to full annotations can lead to improved segmentation robustness. Finally, we will see a method to prioritize slices for segmentation correction to facilitate the annotation process. We will also see the unique challenges of placenta segmentation in MRI and how they can be overcome.
Bella is a PhD candidate at the Hebrew University under the supervision of Prof. Leo Joskowicz. Prior to that, she worked on medical imaging algorithms as a scientist for 7 years at Philips. Involved with the Israeli machine learning community, Bella is also the founder and organizer of the Haifa Machine Learning and Machine Learning for Medical Imaging (MLMI) meetups.