This document summarizes various algorithms that can be used for human organ localization in medical images. It discusses methods such as using probabilistic patch-based label fusion with image registration to localize the left ventricle, global-to-local regression forests to localize multiple organs, and using marginal and full space learning with coarse and fine levels to localize the left ventricle. Additionally, it covers using probabilistic edge maps generated from structured decision forests to localize the left ventricle and atrium, as well as using random decision forests and 3D visual features to detect anatomical structures in CT volumes. The document analyzes approaches such as combining random forests with Hough regression and Markov random fields to localize complex anatomical

1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 11 | Nov -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 230
An Investigation On Human Organ Localization In Medical Images
Anaswara R1, Dhanya Sreedharan2
1Anaswara R: PG Scholar, Dept. of Computer Science and Engineering, SBCE Pattoor, Kerala, India
2Dhanya Sreedharan: Professor, Dept. of Computer Science and Engineering, SBCE Pattoor, Kerala, India
------------------------------------------------------------------------------***---------------------------------------------------------------------------
Abstract - Location and detection of an inner organ in a
medical images is the basic information that is required for
medical image analysis such as image segmentation, lesion
detection, content-based image retrieval, and anatomical
annotation. A general approach/schemeforthelocalizationof
different inner organs that can be adapted to suit various
types of medical image formats is required. This survey
introduces an different algorithms that can be used to solve
organ localization problems.
Keywords: Automatic landmark localization, Regression,
Learning based algorithms, Random forsts.
1. INTRODUCTION
Organ localizationfrommedical imagingmodalities iscrucial
for medical diagnosis such as surgery, needle detection etc.
Organ localization is done by using learning algorithms are
mainly proposed in this survey paper.
The remaining part of the paper is organized as
follows: In Section II survey of all methods will be described
in detail. The paper concludes with a brief summary in
section III.
2. LITERATURE SURVEY
The diagnosis of cardiovascular diseases mainly
deals with the measurement of the leftventricular(LV)mass
and LV cavity volume. This paper proposed magnetic
resonance (MR) image segmentation by multi atlas method.
In this a patch based model[1] is developed ina probabilistic
Bayesian framework and that label information is
incorporated into image registration. Thus a patch based
label fusion model is proposed.
Fast, accurate, and robust localization of several organ is
proposed .Global-to-local cascade of regression forests to
multiple organs are generated. A first regress or encodes
global relationships between organs and subsequent
repressors refine the localization of each organ locally and
independently for improved accuracy. Confidence maps[2]
are used for organ localizationanditgivesinformationabout
the organ shape, regression vote distribution etc.
When an object is subject to large variations and is
surrounded by cluttered background, it is hard to be
detected reliably. In such scenarios, besides the global
detector can train multiple part detectors and use an
intelligent aggregation scheme to exploits the redundancy
among the results of multi detectors to improve robustness.
A novel ranking based scheme was proposed to select the
best LV candidate usingthegeometrical relationshiptoother
candidates. Object localization using marginal space
learning(MSL)[3] and full space learning(FSL) with coarse
and fine are done.
Ultrasound(US) images for cardiovascular diseases must be
combined with tomographic images to provide a road map
for the intervention. Existing multi model US registration
techniques do not achieve reliable registrationduetolow US
image quality. So a probabilistic edgemap(PEM)isproposed
based on the trained decision forests.PEM is generatedfrom
structured decision forests(SDF).PEM[4] focusing on the LV
and atrium while ignoring irrelevantanatomical boundaries.
PEM is a good boundary representation technique. The
boundary taken by PEM is clear and smooth.
A new algorithm for the efficient detection and localization
of anatomical structure with in computer tomography(CT)
volumes are proposed[5].Efficient 3D visual features are
captured by using this system. Those features have been
incorporated within a random decision forest classifier.
Application including efficient visualization and navigation
through 3D medical imagescans.Theoutputofthealgorithm
is probabilistic and therefore enabling the modeling of
uncertainty. This system is applicable to multiple sources of
information(multiple modalities) why because of
probabilistic nature.
An approachforlocalizingcomplex,partlyrepetitive
anatomical structure in 3D volume is proposed which yields
highly accurate robust results with fast runtime.Oursystem
is based on random forests for classification and Hough
regression, the system detects landmark localization. The
model of the anatomical structure is matched to here
landmark candidate by solving Markov Random
Field(MRF)[6].The advantage of the MRF is it does not relay
on predefined intrast point detectorsora manually designed
graph topology but learns both from the training sets. The
combination of random forests andthehoughregressorsare
very important. The MRF is time consuming. They can be
trained on full resolution image data,whiletherandomfores
classifiers can be appliedtostronglydownsampledvolumes.

2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 11 | Nov -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 231
3. CONCLUSION
This survey has been performed for collecting the different
algorithms for organ localization. Localization of inner
organs are very helpful in medical analysis such as sugery,
needle detection .This survey helpsinidentifyingall possible
localization systems for human organ detection and
localization.
ACKNOWLEDGEMENT
I am grateful to my project guide Asst. Prof. Dhanya.
Sreedharan for her remarks, suggestions and for providing
all the vital facilities like providing the Internet access and
important books, which were essential. I am also thankful to
all the staff members of the Department.
REFERENCES
[1] W. Bai et al., “A probabilistic patch-based label fusion
model for multiatlas segmentation with registration
refinement: Application to cardiac MR images,” IEEE Trans.
Med. Imag., vol. 32, no. 7, pp. 1302–1315, Jul. 2013.
[2]R. Gauriau, R. Cuingnet, D. Lesage, and I. Bloch, “Multi-
organ localization combining global-to-local regression and
confidence maps,” in Medical Image Computing Computer-
Assisted Intervent (MICCAI). Springer, 2014, pp. 337–344.
Y. Zheng, X. Lu, B. Georgescu, A. Littmann, E. Mueller, and
[3]D. Comaniciu, “Robust object detection using marginal
space learning and ranking-based multi-detector
aggregation: Application to left ventricledetectionin2DMRI
images,” in Proc. IEEE Comput. Soc. Conf. Comput. Vis.
Pattern Recognit. (CVPR), Jun. 2009, pp. 1343–1350.
[4]O. Oktay et al., “Structured decision forests for multi-
modal ultrasound image registration,” in Proc. Medical
Image Computing Computer-Assisted Intervent (MICCAI).
Springer, 2015, pp. 363–371.
[5]A. Criminisi, J. Shotton,andS.Bucciarelli,“Decisionforests
with longrange spatial context for organ localization in CT
volumes,” in Proc. MICCAI Workshop Probabilistic Models
Med. Image Anal. (PMMIA), Sep. 2009.
[6]R. Donner, B. H. Menze, H. Bischof, and G. Langs, “Global
localization of 3D anatomical structures by pre-filtered
Hough Forests and discrete optimization,” Med. imageAnal.,
vol. 17, no. 8, pp. 1304–1314, 2013.
BIOGRAPHIES
AnaswaraR receivedB.Tech.degree
in Computer Science and
Engineering from Mahathma Gandhi
University, India.Pursuing M.Tech.
degree in Computer Science and
Engineering from Kerala Technical
University, India.
Dhanya sreedharan received B.Tech. Degree in Computer
Science and Engineering from College of Engineering
Karunagappally (CUSAT), received M.Tech. Degree in
Computer and Information Technology from MS University,
India. Currently, she is Assistant Professor, Department of
Computer Science and Engineering, Sree Buddha College
Engineering, Kerala University.
r
Photo