Potential of new 3D ultrasound-based metric to assess the fetal skull: A pilot study

Jacqueline Matthew1, Caroline Knight1, Chandni Gupta1, Alberto Gomez1, Matthew Sinclair2, Yuanwei Li2, Daniel Rueckert2, Juan J Cerrolaza2, 1Department of Perinatal Imaging and Health, King’s College London, 2Biomedical Image Analysis Group, Imperial College London


To evaluate the potential of a novel 3D cranial index (3DCI), derived automatically from 3D ultrasound (US) volumes and to compare 3DCI to the usual method for skull shape assessment  (cephalic index, CI = BPD/OFD).


This retrospective study (NRES ref.num. 14/LO/1806) includes 55 cases (mean gestational age 24.7 weeks, range 20-36) collected during a dedicated US research clinic. All participants had previously had a mid-trimester anomaly scan. Standard 2D scanning planes and 3DUS head volumes were acquired using a Philips Epiq7G scanner with a X6-1 xMatrix transducer. The skull was automatically segmented using a fully-convolutional network architecture, and a statistical model of the normal head shape was generated using principal component analysis and leave-one-out cross-validation. The 3DCI was computed as the distance to the mean shape of the skull normalized to the patient’s gestational age. Additionally, a patient-specific 3D distance map was automatically generated showing in detail the spatial distribution of the distance to the expected shape (see figure attached). The CI was obtained from manually annotated 2D scans. The 5th percentile threshold was used to identify potentially abnormal cases in both metrics. The ground truth for fetal assessment of skull shape was established by a sonographer and a fetal medicine specialist, identifying two cases with dolichocephaly.


The accuracy, specificity and sensitivity for the abnormal shape identification were 90%, 50%, and 92%, for CI, and 98%, 100% and 98% for 3DCI. The new automatic 3DCI significantly outperformed the CI (p-value < 0.005 using McNemar’s statistical test).


The new automatic and objective US-based 3D biometric has the potential to provide objective and accurate assessment of the fetal head, reducing sonographer subjectivity and showing higher diagnostic power than traditional metrics. Finally, the patient-specific morphological map of the fetal skull could allow more objective and quantitative follow up of the patient’s evolution.

View the Poster here