TY - GEN
T1 - Fetal skull segmentation in 3D ultrasound via structured geodesic random forest
AU - Cerrolaza, Juan J.
AU - Oktay, Ozan
AU - Gomez, Alberto
AU - Matthew, Jacqueline
AU - Knight, Caroline
AU - Kainz, Bernhard
AU - Rueckert, Daniel
N1 - Publisher Copyright:
© Springer International Publishing AG 2017.
PY - 2017
Y1 - 2017
N2 - Ultrasound is the primary imaging method for prenatal screening and diagnosis of fetal anomalies. Thanks to its non-invasive and non-ionizing properties, ultrasound allows quick, safe and detailed evaluation of the unborn baby, including the estimation of the gestational age, brain and cranium development. However, the accuracy of traditional 2D fetal biometrics is dependent on operator expertise and subjectivity in 2D plane finding and manual marking. 3D ultrasound has the potential to reduce the operator dependence. In this paper, we propose a new random forest-based segmentation framework for fetal 3D ultrasound volumes, able to efficiently integrate semantic and structural information in the classification process. We introduce a new semantic features space able to encode spatial context via generalized geodesic distance transform. Unlike alternative auto-context approaches, this new set of features is efficiently integrated into the same forest using contextual trees. Finally, we use a new structured labels space as alternative to the traditional atomic class labels, able to capture morphological variability of the target organ. Here, we show the potential of this new general framework segmenting the skull in 3D fetal ultrasound volumes, significantly outperforming alternative random forest-based approaches.
AB - Ultrasound is the primary imaging method for prenatal screening and diagnosis of fetal anomalies. Thanks to its non-invasive and non-ionizing properties, ultrasound allows quick, safe and detailed evaluation of the unborn baby, including the estimation of the gestational age, brain and cranium development. However, the accuracy of traditional 2D fetal biometrics is dependent on operator expertise and subjectivity in 2D plane finding and manual marking. 3D ultrasound has the potential to reduce the operator dependence. In this paper, we propose a new random forest-based segmentation framework for fetal 3D ultrasound volumes, able to efficiently integrate semantic and structural information in the classification process. We introduce a new semantic features space able to encode spatial context via generalized geodesic distance transform. Unlike alternative auto-context approaches, this new set of features is efficiently integrated into the same forest using contextual trees. Finally, we use a new structured labels space as alternative to the traditional atomic class labels, able to capture morphological variability of the target organ. Here, we show the potential of this new general framework segmenting the skull in 3D fetal ultrasound volumes, significantly outperforming alternative random forest-based approaches.
KW - Generalized geodesic distance
KW - Random forest
KW - Structured class
UR - http://www.scopus.com/inward/record.url?scp=85029810507&partnerID=8YFLogxK
U2 - 10.1007/978-3-319-67561-9_3
DO - 10.1007/978-3-319-67561-9_3
M3 - Conference contribution
AN - SCOPUS:85029810507
SN - 9783319675602
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 25
EP - 32
BT - Fetal, Infant and Ophthalmic Medical Image Analysis - International Workshop, FIFI 2017 and 4th International Workshop, OMIA 2017 Held in Conjunction with MICCAI 2017, Proceedings
A2 - Cardoso, M. Jorge
A2 - Arbel, Tal
PB - Springer Verlag
T2 - International Workshop on Fetal and Infant Image Analysis, FIFI 2017 and 4th International Workshop on Ophthalmic Medical Image Analysis, OMIA 2017 held in Conjunction with 20th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2017
Y2 - 14 September 2017 through 14 September 2017
ER -