2168-2194 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/JBHI.2015.2425041, IEEE Journal of Biomedical and Health Informatics
IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, VOL. 11, NO. 4, DECEMBER 2012 1
Standard Plane Localization in Fetal Ultrasound via
Domain Transferred Deep Neural Networks
Hao Chen, Student Member, IEEE, Dong Ni*, Jing Qin*, Shengli Li, Xin Yang, Tianfu Wang
and Pheng Ann Heng, Senior Member, IEEE
Abstract—Automatic localization of the standard plane con-
taining complicated anatomical structures in ultrasound (US)
videos remains a challenging problem. In this paper, we present
a learning based approach to locate the fetal abdominal standard
plane (FASP) in US videos by constructing a domain transferred
deep convolutional neural network (CNN). Compared with pre-
vious works based on low-level features, our approach is able
to represent the complicated appearance of the FASP and hence
achieve better classification performance. More importantly, in
order to reduce the overfitting problem caused by the small
amount of training samples, we propose a transfer learning
strategy, which transfers the knowledge in the low layers of a
base CNN trained from a large database of natural images to our
task-specific CNN. Extensive experiments demonstrate that our
approach outperforms the state-of-the-art method for the FASP
localization as well as the CNN only trained on the limited US
training samples. The proposed approach can be easily extended
to other similar medical image computing problems, which often
suffer from the insufficient training samples when exploiting the
deep CNN to represent high-level features.
Index Terms—Ultrasound, standard plane, deep learning, do-
main transfer, knowledge transfer, convolutional neural network.
I. INTRODUCTION
U
LTRASOUND (US) is a routine screening tool offered to
all pregnant women because of its safety, relatively low
cost and real-time manner. The main goal of a fetal US scan is
to confirm fetal viability, establish gestational age accurately
and look for malformation that could influence prenatal man-
agement. Recent study showed that the sensitivity for prenatal
detection of malformations by US ranges from 27.5% to 96%
in different medical institutes [1]. This wide variation indi-
cates that US-based pregnant diagnosis is operator-dependent
and requires a significant period of training before reaching
competency. Among the pipeline of US diagnosis, acquisition
of the standard plane is the prerequisite step and crucial for
H. Chen is with Department of Computer Science and Engineering,
The Chinese University of Hong Kong, Hong Kong, China (e-mail: jack-
ie.haochen@gmail.com).
D. Ni, J. Qin, X. Yang and T. Wang are with National-Regional Key
Technology Engineering Laboratory for Medical Ultrasound, Guangdong
Key Laboratory for Biomedical Measurements and Ultrasound Imaging,
School of Medicine, Shenzhen University, China (*corresponding author. e-
mail: nidong@szu.edu.cn (D. Ni), jqin@szu.edu.cn (J. Qin)).
S. Li is with Department of Ultrasound, Affiliated Shenzhen Maternal
and Child Healthcare Hospital of Nanfang Medical University, China (e-
mail: lishengli63@126.com).
P. A. Heng is with Department of Computer Science and Engineering,
The Chinese University of Hong Kong and Center for Human Computer
Interaction, Shenzhen Institutes of Advanced Technology, Chinese Academy
of Sciences, China (e-mail: pheng@cse.cuhk.edu.hk).
…
…
… … …
Leg
UC
Heart
SB
UV
SP
Fig. 1: Illustration of the FASP localization from 2-D US
images (the frame with red rectangle is a FASP).
the subsequent biometric measurements and diagnosis [2], [3],
[4]. In this regard, the reported wide sensitivity variation may
be due, at least in part, to the quality of the standard plane
obtained. In clinical practice, acquisition of the standard plane
by clinicians often requires a thorough knowledge of human
anatomy and substantial experience. Therefore, it is very chal-
lenging for novices and even difficult and time consuming for
clinical experts. Thus, the development of automatic methods
for locating standard planes from 2-D US images would assist
the novices as well as improve the efficiency of experts.
In this study, we focus on automatically locating the fe-
tal abdominal standard plane (FASP) from US videos (a
preliminary version of this work has been reported in [5]
and significant improvements have been made to the original
paper). Clinically, to locate the FASP, a radiologist attempts to
find the concurrent presence of three key anatomical structures
(KASs): the stomach bubble (SB), the umbilical vein (UV) and
the spine (SP) in one frame when moving the US probe across
the patient body. The procedure is illustrated in Fig. 1. Based
on the acquired FASP, the clinician can measure abdominal
circumference (AC), which is the most important measurement
for estimating fetal weight. The accuracy of AC measurement
is heavily dependent on both the quality of the FASP and
the manual measurement on the FASP by clinicians. Recently,
commercial tools have been developed for the automatic
AC measurement on several US scanners including Siemens
Acuson S2000, GE LOGIQ S8, Mindray DC8, etc. However,
little attention has been paid to the prerequisite step, that is,
FASP acquisition.
Over the past few years, some methods have been proposed
for locating standard planes from 2-D US images. Zhang