More reliable diagnoses of fetal anomalies through 3D reconstruction of the maternal uterus

Magnetic resonance imaging (MRI) is especially useful for examining fetuses for suspected abnormalities on ultrasound, but the technique is traditionally limited by maternal-fetal movement during image acquisition.

Now, a technique being worked on by researchers at the Polytechnic University of Madrid (UPM) in Spain could enable a game changer in this field, by making it possible to obtain detailed 3D reconstructions of the entire maternal uterus from continuous imaging of a moving fetus by MRI.

This is possible by combining advanced models of maternal-fetal movement and unsupervised deep learning techniques. The study was carried out by researchers from the UPM Biomedical Imaging Technologies Group, led by Lucilio Cordero Grande and coordinated by María Jesús Ledesma Carbayo, with the collaboration of King’s College London in the United Kingdom, the Biomedical Research Center in the Bioengineering, Biomaterials and Nanomedicine Network (CIBERBBN) in Spain and the Gregorio Marañón General University Hospital in Madrid, Spain.

“Due to its simplicity, ultrasound is the gold standard imaging modality for pregnancy monitoring and is therefore commonly associated with visualization of fetal development. However, magnetic resonance imaging is the modality indicated when ultrasound anomalies are suspected,” explains Lucilio Cordero, project researcher and professor at the UPM’s Higher Technical School of Telecommunications Engineers (ETSIT). Because of the long acquisition time and maternal-fetal movement, “until now, consistent representations of the entire fetal anatomy have not been able to be constructed from MRI. Thus, previous studies had focused especially on the fetal brain or torso separately”, he adds.

On the left, the result of a conventional fetal MRI is shown on three image slices (red, green and yellow). Inconsistencies due to fetal movement are observed. In the center are the same plans reconstructed with the proposed technique. Better resolution and consistent 3D information is observed. On the right, the surface of the fetal body obtained by means of a fully automatic algorithm, developed by doctoral student Pedro Pablo Alarcón Gil as part of the research, is shown superimposed. (Images: Pedro Pablo Alarcón Gil / UPM)

The current proposal by the UPM researchers breaks with these limitations and allows the development of a technique for 3D fetal image reconstruction using MRI with complete coverage of the maternal uterus. The results show the possibility of obtaining high-quality three-dimensional representations with complete coverage of fetal and uterine tissues.

“The comparison with existing methods, which only attend to 3D consistency in localized regions such as the fetal brain or torso, has shown that, in addition to being applicable to a larger field of view, the proposed technique provides a significant increase in resolution and image quality”, indicates María Jesús Ledesma, professor at ETSIT.

The results obtained in this project have been validated by radiologists who are experts in fetal imaging and the added value provided in applications aimed at clinical problems has been demonstrated.

By enabling a new way to build robust and detailed representations of fetal anatomy, the future applications of this work could be numerous. For now, improved image quality has resulted in greater reliability in the use of machine learning techniques to obtain biomarkers of potential clinical interest, such as estimated gestational age.

In addition, the possibility of 3D panoramic visualization of the fetal body has allowed the direct and objective quantification of fetal growth. In the latter case, a high level of agreement has been observed with population results obtained from indirect growth assessment methods using ultrasound, which require an expert operator and adequate conditions for correct visualization of the fetus.

Through this research, preliminary results have also been obtained that point to the interest of the technique in other clinical domains, and its use to improve ultrasound images is even being studied. Among the objectives of this project is, as explained by its researchers, to explore and promote potential advances in clinical management based on the new technique.

Cordero Grande, Ledesma Carbayo and their colleagues present the technical details of their research in the academic journal IEEE Transactions on Medical Imaging, under the title “Fetal MRI by robust deep generative prior reconstruction and diffeomorphic registration”. (Source: UPM)