Eye Gaze Estimation Invisible and IR Spectrum for Driver Monitoring System

Signal & Image Processing: An International Journal (SIPIJ) Vol.11, No.5, October 2020
DOI: 10.5121/sipij.2020.11502 21
BATCH NORMALIZED CONVOLUTION NEURAL
NETWORK FOR LIVER SEGMENTATION
Fatima Abdalbagi1
, Serestina Viriri2
and Mohammed Tajalsir Mohammed3
1
Department of Computer Science, Sudan University of Science
and Technology, Khartoum, Sudan
2
Computer Science Discipline, University of KwaZulu-Natal, Durban, South Africa
3
Department of Computer Science, Sudan University of Science
and Technology, Khartoum, Sudan
ABSTRACT
With the huge innovative improvement in all lifestyles, it has been important to build up the clinical fields,
remembering the finding for which treatment is done; where the fruitful treatment relies upon the
preoperative. Models for the preoperative, for example, planning to understand the complex internal
structure of the liver and precisely localize the liver surface and its tumors; there are various algorithms
proposed to do the automatic liver segmentation. In this paper, we propose a Batch Normalization After All
- Convolutional Neural Network (BATA-Convnet) model to segment the liver CT images using Deep
Learning Technique. The proposed liver segmentation model consists of four main steps: pre-processing,
training the BATA-Convnet, liver segmentation, and the postprocessing step to maximize the result
efficiency. Medical Image Computing and Computer Assisted Intervention (MICCAI) dataset and 3D-
Image Reconstruction for Comparison of Algorithm Database (3D-IRCAD) were used in the
experimentation and the average results using MICCAI are 0.91% for Dice, 13.44% for VOE, 0.23% for
RVD, 0.29mm for ASD, 1.35mm for RMSSD and 0.36mm for MaxASD. The average results using 3D-
IRCAD dataset are 0.84% for Dice, 13.24% for VOE, 0.16% for RVD, 0.32mm for ASD, 1.17mm for
RMSSD and 0.33mm for MaxASD.
KEYWORDS
Liver Segmentation, Convolutional Neural Network, Computed Tomography
1. INTRODUCTION
Usually, the liver segmentation process done manually by trained clinicians but it is time
consuming and requiring much effort and also it different from one clinician to another because
of the observer variability; as the result of that, an automatic liver segmentation system would
thus be a great boon for perform these tasks. Because of complexity of liver shapes and variable
liver sizes among patients the segmentation of the liver from medical images is very difficult and
also due to low contrast between the liver and surrounding organs like stomach, pancreas, kidney
and muscles. The liver is one of the biggest and an essential organ in the human body. It is
molded like a cone and situated in the upper right-hand part of the stomach cavity, underneath the
stomach. It is responsible for carrying out some very important functions to keep the body pure of
toxins and harmful substances [1]. In excess of 500 real capacities have been related to the liver
and some of them remember guideline of most compound levels for the blood, discharge of bile
and separating blood that originates from the stomach related parcel [2]. Modern surgeries rely on
Computer Aided Diagnosis (CAD) systems to assist doctors in the diagnosis of medical images
and surgical planning. CAD it is one of the major research topics because it is part of the routine

22
clinical work in medical imaging and diagnostic radiology [3]. Liver pathologies such as
cirrhosis, liver cancer and fulminant hepatic failure can be diagnosed by using non-invasive
techniques (Medical imaging) such as computed tomography (CT), Magnetic Resonance Imaging
(MRI), Ultrasound (US), Positron Emission Tomography (PET) or Single-Photon Emission
Computed Tomography (SPECT). One of the advantages of CT images is their higher signal to
noise ratio, better spatial resolution and they provide more accurate anatomical information about
the visualized structures; therefor, the diagnosticians are preferred to use a CT image [4].
Accurate, efficient and automatic methods for liver segmentation are demanded because
traditionally the radiologists and physicians have to manually delineate the liver region slice by
slice, which is tedious and time consuming due to the large amount of data [5]. There are several
clinical applications for automatic liver segmentation such as, measuring the graft volume before
living donor liver transplantations [6], locating vessels and tumors, it is also useful to monitor
patients with liver metastases, where the disease is related to an enlargement of the liver, and it
plays an important role in surgery planning which it is most time consuming step [7]. Information
about precise size for liver, localize the tumors and accurate liver surface segmentation are
required in liver local treatment [8]. Due to unpredictability of liver shapes and variable liver
sizes among patients the segmentation of the liver from medical images is troublesome because
of low differentiation between the liver and encompassing organs like stomach, pancreas, kidney
and muscles [9]. Moreover, challenge is presence large tumors and other liver pathologies
because the livers with pathologies are different from healthy ones and that results either on
segmentation or over segmentation. In the last few decades, a large variety of semiautomatic and
fully automatic approaches have been proposed to improve the liver segmentation procedure,
such as: region growing, clustering, deformable models or level sets, statistical shape models
(SSMs), probabilistic atlases, graph cuts and recently, deep learning techniques [5] [10].
As of late, deep-learning strategies have been utilized in the field of medical image processing
[8]. There are three principles building blocks of deep learning methods. These are Convolutional
Neural Network (convnet), Auto Encoder (AE) and Deep Belief Network (DBN). These deep
learning architectures as of now have been applied in numerous fields like bio-informatics,
speech, audio recognition and computer vision [9]. In this paper we focus on using deep learning
techniques to detect and segment the liver and compare the result of proposed model with some
of the literature.
The ConvNet has gained significant attention for solving computer vision tasks such as object
recognition, classification and segmentation [11] [12]. The proposed method is considered as a
binary labelling problem which is assigning the labels to each voxel in original image into either
belonging to liver or not, and it will be carried out by using convolutional neural network;
resulting on a probability map which identifies the liver surface. The contribution made by this
paper is the use of a deep learning technique to segment the liver with new concept which adding
a BatchNormalization layer after each convolution layer.
The rest of this paper is organized as follows: the literature review and related work are provided
in section 2 and section 3 brieﬂy describes the methods and techniques used in this work. The
results are detailed and discussed in section 4. Finally, in section 6 the conclusions are drawn.
2. LITERATURE REVIEW AND RELATED WORK
There are many semiautomatic and completely automatic approaches have been proposed to
improve the liver segmentation methodology. A new method was proposed in [10], for automatic
segmentation of liver using supervoxel based graph cuts. Further in G. Litjens et al. (2017)

23
mentioned that the new method automatically extracts the Liver Volume of Interest (VOI) and
the foreground/background seed points for graph cuts. Firstly, they determined the region of
abdomen by using the Maximum Intensity Projection (MIP) and thresholding methods. And
extract the specific liver VOI from the region of abdomen according to prior knowledge about
locating liver organ and by using a histogram based adaptive thresholding method and
morphological operations. They generated the supervoxels of the liver VOI by using the Simple
Linear Iterative Clustering (SLIC) method. Besides, foreground/background seeds for graph cuts
were chosen on the largest liver slice, and the graph cuts algorithm was applied to the VOI
supervoxels.
A Fully automatic CT liver segmentation using a novel statistical shape model approach was
presented by [13], they combine learned local shape constraints with observed shape deviation
during adaptation. This study [14] introduced a CNN for liver segmentation on abdominal
computed tomography (CT) images with focusing on the performance of generalization. They
show high generalization of performance and accuracy. They proposed an auto-context neural
network; it achieved an effective estimation to obtain the shape prior. They use a self-supervised
contour scheme to extend their network. They achieved better accuracy when compared to the
state-of-the-art networks by reducing 10.31% of the Hausdorff distance.
A novel 3D deeply supervised network (DSN) for automatic liver segmentation was presented by
[15]. This method employs a fully convolutional architecture to produce a high-quality score map
which is processed further by the employment of a conditional random field (CRF) to obtain a
refined segmentation. The MICCAI-SLiver07 dataset was used. This method was trained on
twenty CT scans and test on ten CT scans. Although the 3D DSN generates high-quality
probability maps, a graphical model is employed to re ne segmentation results due to imprecise
segmentation in ambiguous regions. A fully connected CRF model on the transverse plane is
exploited which solves energy function to re ne the segmentation. The segmentation results
achieved were a VOE of 5.42% and an average symmetric surface distance of 0.79mm.
The authors in [16] proposed a new 3D liver segmentation method based on multi-region
appearance and graph cuts approach in order to reducing user interaction and improving the
accuracy and efficiency. The liver could contain tumor or metastasis which results liver with
multiple sub-regions, for such case they introduced a novel multiregion-appearance model and
appearance selection scheme to segmenting the target multiregion object. The graph cuts
approach was used to optimize the proposed energy function. They had compared their work with
other graph cuts based methods, state-of-the-art semiautomatic and interactive methods and with
prior model based methods. They found that the proposed model needs only initial seeds in the
liver when compared with the other graph cuts methods and it required low interactive compared
with the other semiautomatic and interactive methods and also the proposed model can be applied
to livers with any shape because it was not restricted by specific training data when compared
with prior model based methods.
In [17], a fully automatic framework was proposed for liver segmentation based on 3D
convolutional neural network (CNN) and globally optimized surface evolution. Firstly, the deep
3D CNN gives the initial liver surface after it was trained to learn a subject-specific probability
map of the liver. Then, refining the initial liver segmentation by using the prior information about
novel energy function; Finally, propagated the initial liver surface to the optimal position by
minimized the energy function using a global optimization-based approach. A semi-automatic
method to identify tumors from 3D CT scans based on 2D region growing with knowledge-based
constraints was proposed by [18]. Firstly, they decompose the 3D scan image into component
slices. Then, they apply 2D region growing with knowledge-based constraints on each slice.

24
Finally, they load up the individual segmented lesions together to generate a 3D volume. The
region-growing approach used to calculate the seed point and feature vectors and also to label the
voxels. The Knowledge-based constraints were used to ensure that the segmented region size and
shape is within acceptable parameters. The method required a minimal user involvement in order
to define an approximate region of interest around the lesion in each slice image which improved
the performance of region growing, as well as reduces computational requirements.
An application based of minimal surfaces and Markov random fields to segmentation the liver
tumors were offered in [19]. They were using a region graph instead of a pixel graph in applying
these models which leads to an interactive method used to segmenting tumors in 3D CT. their
procedure was rely upon the manual definition of a sub-volume containing one or more tumors
that need to be segmented based on simple observation which is: the liver presents two types of
tissues: tumoral and healthy tissues to classification the liver pixels they were model the liver
pixels as a Markov Random Field and the classification is performed through the maximum a
posteriori estimation. But this classification step is supervised by user in defining the markers that
specify both tumoral and healthy tissues. These markers are used to locate the tumors and to
estimate the grey levels characteristics of these structures. They compared their work with the
radiologist's segmentations on a set of 5 CT images presenting 10 tumors and found that the mean
surface distance between their segmentations and the radiologists is approximately 1.5 mm.
approximately 71 % of their segmentations is in perfect match with the radiologist's
segmentation. The total mean score obtained on the training data set was equal to 88. And the
computation time needed for the tumor segmentation is approximately equal to five, up to height
minutes.
Christ et al. ware presented a combined automatic segmentation of the liver and its lesions in CT
and MRI abdomen images using two cascaded fully convolutional neural networks (CFCNs) one
for the segmentation of the liver and the other for its lesions [20]. They use an abdominal CT
dataset comprising 100 hepatic tumor volumes for training the CFCN models. In the first step of
their work they preprocessed the CT/MRI image with HU-windowing or N4 bias correction.
Then in second step, segmenting the liver from abdomen CT/MRI scans using first pertained
FCN after that they use the output of last step to be input for a second FCN which segments
lesions from the given segmented liver ROI. And finally last step was the post-processing using
3D Conditional Random Field (3D CRF). Their results show that the CFCN achieves Dice scores
over 94% for the liver with computation times below 100s per volume. And in contrast to prior
work their method could be generalized to segment multiple organs in medical data using
multiple cascaded FCNs.
A literature review of medical image segmentation based on U-net was presented by [21]. They
were focused on the successful segmentation experience of U-net in six medical imaging systems
including computed tomography (CT), magnetic resonance imaging (MRI), ultrasound, X-ray,
optical coherence tomography (OCT), and positron emission computed tomography (PET). There
are many kinds of lesion regions extracted by these applications. Also in this study, they were
introduced the method of combining the original U-net architecture with deep learning and a
method for improving the U-net network. They came out that the six imaging systems mentioned
in their article, are not perfect in some imaging systems. Consequently, it needs to be improved in
future studies for application in various imaging systems.
Two deep encoder-decoder convolutional neural networks (EDCNN) were constructed and
trained to cascade segments of both the liver and lesions in CT images in this paper [22]. The first
EDCNN was responsible for segmented the liver image which has been an input for the training
of a second EDCNN; where it then responsible for segments the tumor regions within the liver

25
ROI regions as predicted by the ﬁrst EDCNN. In order to evaluate their performance, the
proposed EDCNN networks produced an average DICE score of 95.22% for the test set of CT
images using a public dataset (3DIRCADb) and it was compared with some of the existing
methods. The experimental results demonstrated that the proposed EDCNN achieved improved
performance in segmentation accuracy over some existing methods.
The problem of liver segmentation is extremely difficult thus there is no one good solution till
now. There are many whose have presented some of the most popular techniques which attempt
to deal with this problem like in [35] although they produce reasonable segmentations in many
situations but at some point local ambiguities and errors introduced by the segmentation process.
And more work needs to be done to match the human detection performance [36]. Still some
methods have under-segmentation or over-segmentation errors [37]. These errors need to be
reduced in order to improve the accuracy of the segmentation process and that what we try to do
in this paper.
The liver segmentation is very complicated so we had to make use of some of the new techniques
presented here in this paper. We propose a model based on deep learning technique. Recently,
deep learning has achieved state-of-the-art performance in medical image analysis and have been
applied to medical image analysis with the intention of letting computers learn the features that
optimally represent the problem data at hand [38].
3. METHODS AND TECHNIQUES
Liver segmentation model is implemented in this paper using Batch Normalization Convolutional
Neural Network (BATA-Convnet). Our Methodology for Liver segmentation consists of four
steps as shown in the flowchart in Figure 1. First, pre-processing the CT image then, training the
convnet Architecture. Third step is the liver segmentation. Last step is post processing step to
maximize the result efficiency.
3.1. Datasets
The datasets used in this work are the Medical Image Computing and Computer Assisted
Intervention (MICCAI) 2007 grand challenge dataset and 3D Image Reconstruction for
Comparison of Algorithm Database (3D-IRCAD). All CT images in MICCAI are enhanced with
different operator and scanned in the focal venous stage on an assortment of scanners (different
makers, 4, 16 and 64 detector lines). As it is CT, all datasets were acquired in cross-over
direction. The pixel dispersing varied somewhere in the range of 0.55 and 0.80 mm and inter-
slice distance varied from 1 to 3 mm. The majority of the images is pathologic and includes
tumors, metastasis and sores of different sizes. The dataset used for preparing is the MICCAI
2007 grand challenge preparing set (MICCAI-Training) which comprises of 20 volume images
with corresponding ground truth segmentations. The dataset used for testing is the MICCAI 2007
grand challenge testing data (MICCAI-Testing) which comprises of 10 volume images. The 3D-
IRCAD database is composed of the CT scan of 20 patients (10 men and 10 women) with hepatic
tumours in 75% of cases. The images are provided by the authors in DICOM and VTK format in
512x512 pixels. It also contains handmade true segmentation for liver, bones, tumors and others
by medical specialists for all images of the 20 patients. The CT image in 3D-IRCAD is 512x512,
but may take a very long time to complete training such as images on personal computer, so the
images were scaled down to 128x128 pixels.

26
Figure.1 The flowchart of proposed model.
3.2. Pre-Processing
In spite of the fact that convnet require signiﬁcantly less image preprocessing than customary
techniques, this is as yet a fundamental errand that can improve results. The CT volumes included
the crude information of every clinical focus, and thusly, may contain commotion. To ﬁlter such
unfortunate data, CT Windowing method was utilized.
Windowing is also known as contrast stretching, grey-level mapping, histogram modification or
contrast enhancement is a significant methodology in a CT scan which is the process in which the

27
CT image Grayscale component of an image is manipulated via the CT numbers, Pixel intensity
measured in Hounsfield Units (HU). Depending on many experimental Sahi et al [24] propose to
use the range [-62, 238] HU and P. F. Christ et al. was truncated the image intensity of all CT
scans to the range [-100, 400] HU [25]. As such, our CT scans range was set to the range [-150,
230] HU as shown in Figure 2.
Figure.2 CT Windowing preprocessing
3.3. Liver Segmentation
The architecture of the proposed BATA-Convnet as shown in Figure.3 which it includes twelve
conv layers (convolution). All convolution layers are followed by a Batch Normalization layer
also it's followed by a rectified linear unit (ReLU) to speed up the training. Batch Normalization
layer is used to normalizes the values going into each activation function by applying a
transformation that maintains the mean output close to zero and the output standard deviation
close to one [39]. The network takes an input image of size 128 x 128 which corresponds to
feature maps of a 2D slice of the CT volume.
Figure.3 The Architecture of Proposed model.
3.4. Post Processing
Post-processing is used to refine the segmentation results produced by the previous segmentation
step, including thresholding, largest connected component selection (LCCS) and morphological
operations to filling the hole.
3.4.1. Thresholding
The Binary Thresholding is used to eliminate the pixels with a probability below 50 percent in
the probability map, which in our case is a pixel value 127 which is 50% of the maximum pixel
value of 255 to do that we use an cv2.threshold() function in OpenCV library. The result of the

28
thresholding post processing can be seen in Figure 4 where 4(a) and 4(b) are slices before and
after thresholding, respectively.
(a) (b)
Figure.4 The Result of the thresholding.
3.4.2. Largest Connected Component Selection (Lccs)
The Largest connected component selection (LCCS) technique has been used as a primary
segmentation or as post-processing. In our case we used it as post-processing to refine the final
result. We calculate the Largest connected component in resulting segmentation volume by
setting the background pixel intensity to zero and each component that is not zero within the
volume is labeled. Then using python Image Processing Toolbox to calculate the volume of each
component and the component with the largest volume is selected. This volume is then set as the
threshold volume and all components with a volume less than this threshold volume are removed
using morphological operators, this is done using remove_small_objects () function in the image
processing Python package scikit-image. The result of the LCCS can be seen in Figure. 5 where
5(a) and 5(b) are slices before and after LCCS process, respectively.
(a) (b)
Figure.5 The result of the LCCS process.
3.4.3. Filling the Hole
After the largest connected component is detected and the smaller components are removed, the
resulting volume is processed using Morphological closing to remove the hole from it by using

29
remove_small_holes() function in scikit-image. The result of the holes filling process can be seen
in Figure 6 where 6(a) and 6(b) are slices before and after the process, respectively.
(a) (b)
Figure.6 The result of the Holes Filling process.
4. RESULTS AND DISCUSSION
Our proposed BATA-Convnet model for Liver Segmentation consist of twelve conv layers
(convolution), two maxpooling layers, two upsampling layers, two concatenate layers as well as a
softmax layer. It's implemented in laptop with processor Intel Core(TM) i7-6500U and CPU @
2.59 GHz with 8.00 GB RAM and 64 bit operating system. The implementation is done using
Python Programming Language and OpenCV library for programming functions with Keras and
Tensorflow open-source neural-network library for the implementation of deep learning
techniques.
4.1. Experimental Setup
An experiment was set up to test BATA-Convnet model for Liver Segmentation in a scientific
manner. The model was trained and validated on MICCAI and 3D-IRCAD datasets described
above. The network is trained using 75% of training and the rest for validation. The Training and
validation loss and accuracy curves for MICCAI and 3D-IRCAD datasets are depicted in
Figure.7 and Figure.8, respectively.
Figure.7 The Training, validation loss and accuracy curves for MICCAI dataset.

30
Figure.8 The Training, validation loss and accuracy curves for 3D-IRCAD dataset.
During each iteration128 x 128 slices is randomly chosen from the training data and 128 x 128
slices from the ground truth slices as the output. Training the network takes approximately 45
hours using the laptop described above. During the training the validation errors or loss are used
to evaluate the network. After training the model, the networks are used to segment the liver in
the 2D slices and the probability maps obtained for MICCAI and 3D-IRCAD datasets are
depicted in Figure 9 and Figure 10, respectively. Where Figure 9(a), 9(b) and 9(c) are the original
images and 9(d), 9(e) and 9(f) are the probability maps obtained by BATA-Convnet model,
respectively. And Figure 10(a), 10(b) and 10(c) are the original images and 10(d), 10(e) and 10(f)
are the probability maps obtained by BATA-Convnet model, respectively.
Figure.9 Sample of probability maps from MICCAI dataset.

31
Figure.10 Sample of probability maps from 3D-IRCAD dataset.
Subsequently, the resulting segmented volumes are compared to the ground truth volumes using
some evaluation metrics such as: volume overlap error (VOE), relative volume difference (RVD),
maximum symmetric surface distance (Max-ASD), average symmetric surface distance (ASD)
and root mean square symmetric surface distance (RMSSD)[26]. The average results using
MICCAI are 0.91% for Dice, 13.44% for VOE, 0.23% for RVD, 0.29mm for ASD, 1.35mm for
RMSSD and 0.36mm for MaxASD. The average result using 3D-IRCAD dataset are 0.84% for
Dice, 13.24% for VOE, 0.16% for RVD, 0.32mm for ASD, 1.17mm for RMSSD and 0.33mm for
MaxASD. Thereafter, the result was compared to some related literature [16] [19] [27] [28] [29]
[30] [31] [32] [33] [34] as shown in table I. The results in terms of volume overlap error were
better than those results [28] [19] [31] as shown in Figure 11 and in terms of the relative volume
difference were better than those results [27] [30] [31] as shown in Figure 12. But in in terms of
(ASD), (RMSSD) and (Max-ASD) our result is better than all the other results as shown in Figure
13, Figure 14 and Figure 15, respectively
Table 1: Comparison to related literature
Related Work VOE RVD ASD RMSSD Max-ASD
[16] 4.53 0.15 0.79 1.49 16.75
[27] 7.87 1.31 1.29 2.50 23.56
[28] 72.38 - - - -
[19] 29.49 - - - -
[29] 5.36 0.03 0.96 1.84 19.20
[30] 5.90 2.70 0.91 1.88 18.94
[31] 16.75 2.05 - - -
[32] 5.42 - 0.79 - -
[33] - - - - -
[34] - - - - -
Result for
MICCAI
13.44% 0.23% 0.29mm 1.35mm 0.36mm
Result for 3D-
IRCAD
13.24% 0.16% 0.32mm 1.17mm 0.33mm

32
Figure.11 The results in terms of VOE
Figure.12 The results in terms of RVD
Figure.13 The results in terms of ASD

33
Figure.14 The results in terms of RMSSD
Figure.15 The results in terms of Max-ASD
5. CONCLUSION
The BATA-Convnet model for liver segmentation using 3D CT scans is proposed in this paper.
After trying many experiments for the architecture of our network we come out to propose a
BATA-Convnet, the ordinary Convnet it's stack of the convolution followed by maxpooling
layers here we proposed to add Batch Normalization layer after all convolution layer which
enhance the overall accuracy. BATA-Convnet model is implemented using Python Programming
Language and OpenCV library for programming functions with Keras and TensorFlow Open-
Source Neural-Network library. MICCAI and 3D-IRCAD datasets were used in the
experimentation and the average results using MICCAI are 0.91% for Dice, 13.44% for VOE,
0.22% for RVD, 0.29mm for ASD, 1.35mm for RMSSD and 0.36mm for MaxASD. The average
result using 3D-IRCAD dataset are 0.84% for Dice, 13.24% for VOE, 0.16% for RVD, 0.32mm
for ASD, 1.17mm for RMSSD and 0.33mm for MaxASD.
Conflicts of Interest
The authors declare no conflict of interest.
REFERENCES
[1] V. Dixit, J. Pruthi, “Review of Image Processing Techniques for Automatic Detection of Tumor in
Human Liver,” International Journal of Computer Science and Mobile Computing, 33(3), 371–378,
2014.
[2] Johns Hopkins Medicine, 2019.
[3] K. Doi, “Computer-aided diagnosis in medical imaging: Historical review, current status and future
potential,” Computerized Medical Imaging and Graphics, 31(4–5), 198–211, 2007,
doi:10.1016/j.compmedimag.2007.02.002.

34
[4] P. Campadelli, E. Casiraghi, A. Esposito, “Liver segmentation from computed tomography scans: A
survey and a new algorithm,” Artificial Intelligence in Medicine, 45(2–3), 185–196, 2009,
doi:10.1016/j.artmed.2008.07.020.
[5] Y. Nakayama, Q. Li, S. Katsuragawa, et al., “Automated hepatic volumetry for living related liver
transplantation at multisection CT.,” Radiology, 240(3), 743–748, 2006,
doi:10.1148/radiol.2403050850.
[6] L. Hermoye, I. Laamari-Azjal, Z. Cao, L. Annet, J. Lerut, B.M. Dawant, B.E. Van Beers, “Liver
Segmentation in Living Liver Transplant Donors: Comparison of Semiautomatic and Manual
Methods,” Radiology, 234(1), 171–178, 2005, doi:10.1148/radiol.2341031801.
[7] T. Heimann, I. Wolf, H.-P. Meinzer, “Active shape models for a fully automated 3D segmentation of
the liver--an evaluation on clinical data.,” Proc. Medical Image Computing and Computer-Assisted
Intervention, 9(Pt 2), 41–8, 2006, doi:10.1007/11866763_6.
[8] L. Massoptier, S. Casciaro, “A new fully automatic and robust algorithm for fast segmentation of
liver tissue and tumors from CT scans,” European Radiology, 18(8), 1658–1665, 2008,
doi:10.1007/s00330-008-0924-y.
[9] X.Z.X. Zhang, J.T.J. Tian, K.D.K. Deng, Y.W.Y. Wu, X.L.X. Li, A.S. Construction, “Automatic
liver segmentation using a statistical shape model with optimal surface detection,” Biomedical
Engineering, IEEE Transactions on, 57(10), 2622–2626, 2010, doi:10.1109/TBME.2010.2056369.
[10] G. Litjens, T. Kooi, B.E. Bejnordi, A.A.A. Setio, F. Ciompi, M. Ghafoorian, J.A.W.M. van der Laak,
B. van Ginneken, C.I. Sánchez, “A Survey on Deep Learning in Medical Image Analysis,”
(February), 2017, doi:10.1016/j.media.2017.07.005.
[11] J. Long, E. Shelhamer, T. Darrell, “Fully convolutional networks for semantic segmentation,”
Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,
07–12–June, 3431–3440, 2015, doi:10.1109/CVPR.2015.7298965.
[12] A. Krizhevsky, G.E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,”
1–9.
[13] M. Erdt, S. Steger, M. Kirschner, S. Wesarg, “Fast automatic liver segmentation combining learned
shape priors with observed shape deviation,” Proceedings - IEEE Symposium on Computer-Based
Medical Systems, 249–254, 2010, doi:10.1109/CBMS.2010.6042650.
[14] M. Chung, J. Lee, J. Lee, Y.-G. Shin, “Liver Segmentation in Abdominal CT Images via Auto-
Context Neural Network and Self-Supervised Contour Attention,” 2020.
[15] Q. Dou, H. Chen, Y. Jin, L. Yu, J. Qin, P.A. Heng, “3D deeply supervised network for automatic
liver segmentation from CT volumes,” Lecture Notes in Computer Science (Including Subseries
Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 9901 LNCS, 149–157,
2016, doi:10.1007/978-3-319-46723-8_18.
[16] J. Peng, P. Hu, F. Lu, Z. Peng, D. Kong, H. Zhang, “3D liver segmentation using multiple region
appearances and graph cuts,” Medical Physics, 42(12), 6840–6852, 2015, doi:10.1118/1.4934834.
[17] P. Hu, F. Wu, J. Peng, P. Liang, D. Kong, “Automatic 3D liver segmentation based on deep learning
and globally optimized surface evolution,” Physics in Medicine and Biology, 61(24), 8676–8698,
2016, doi:10.1088/1361-6560/61/24/8676.
[18] D. Wong, J. Liu, Y. Fengshou, Q. Tian, W. Xiong, J. Zhou, Y. Qi, T. Han, S.K. Venkatesh, S. Wang,
“A semi-automated method for liver tumor segmentation based on 2D region growing with
knowledge-based constraints,” The Midas Journal, (January), 2008.
[19] J. Stawiaski, E. Decencière, F. Bidault, J. Stawiaski, E. Decencière, F. Bidault, I. Liver, T.
Segmentation, “Interactive Liver Tumor Segmentation Using Graph-cuts and Watershed To cite this
version : HAL Id : hal-01445751 Graph-cuts and Watershed,” 2017.
[20] P.F. Christ, F. Ettlinger, F. Grün, M.E.A. Elshaera, J. Lipkova, S. Schlecht, F. Ahmaddy, S.
Tatavarty, M. Bickel, P. Bilic, M. Rempfler, F. Hofmann, M.D. Anastasi, S.-A. Ahmadi, G. Kaissis,
J. Holch, W. Sommer, R. Braren, V. Heinemann, B. Menze, “Automatic Liver and Tumor
Segmentation of CT and MRI Volumes using Cascaded Fully Convolutional Neural Networks,” 1–
20, 2017.
[21] G. Du, X. Cao, J. Liang, X. Chen, Y. Zhan, “Medical image segmentation based on U-Net: A
review,” Journal of Imaging Science and Technology, 64(2), 1–12, 2020,
doi:10.2352/J.ImagingSci.Technol.2020.64.2.020508.

35
[22] Ü. Budak, Y. Guo, E. Tanyildizi, A. Şengür, “Cascaded deep convolutional encoder-decoder neural
networks for efficient liver tumor segmentation,” Medical Hypotheses, 134(September 2019), 2020,
doi:10.1016/j.mehy.2019.109431.
[23] S. Luo, X. Li, J. Li, “Review on the Methods of Automatic Liver Segmentation from Abdominal
Images,” Journal of Computer and Communications, 2(2), 1–7, 2014, doi:10.4236/jcc.2014.22001.
[24] K. Sahi, S. Jackson, E. Wiebe, G. Armstrong, S. Winters, R. Moore, G. Low, “The value of ‘liver
windows’ settings in the detection of small renal cell carcinomas on unenhanced computed
tomography,” Canadian Association of Radiologists Journal, 65(1), 71–76, 2014,
doi:10.1016/j.carj.2012.12.005.
[25] P.F. Christ, F. Ettlinger, F. Grün, M.E.A. Elshaera, J. Lipkova, S. Schlecht, F. Ahmaddy, S.
Tatavarty, M. Bickel, P. Bilic, M. Rempfler, F. Hofmann, M.D. Anastasi, S.-A. Ahmadi, G. Kaissis,
J. Holch, W. Sommer, R. Braren, V. Heinemann, B. Menze, “Automatic Liver and Tumor
Segmentation of CT and MRI Volumes using Cascaded Fully Convolutional Neural Networks,”
(February), 2017.
[26]. M. Styner, J. Lee, B. Chin, M. Chin, “3D segmentation in the clinic: A grand challenge II: MS lesion
segmentation,” Midas, 1–6, 2008.
[27] W. Wu, Z. Zhou, S. Wu, Y. Zhang, “Automatic Liver Segmentation on Volumetric CT Images Using
Supervoxel-Based Graph Cuts,” Computational and Mathematical Methods in Medicine, 2016(April),
1–14, 2016, doi:10.1155/2016/9093721.
[28] M. Goryawala, S. Gulec, R. Bhatt, A.J. McGoron, M. Adjouadi, “A Low-Interaction Automatic 3D
Liver Segmentation Method Using Computed Tomography for Selective Internal Radiation Therapy,”
BioMed Research International, 2014(1), 1–12, 2014, doi:10.1155/2014/198015.
[29] P. Hu, F. Wu, J. Peng, P. Liang, D. Kong, “Automatic 3D liver segmentation based on deep learning
and globally optimized surface evolution,” Physics in Medicine and Biology, 61(24), 8676–8698,
2016, doi:10.1088/1361-6560/61/24/8676.
[30] F. Lu, F. Wu, P. Hu, Z. Peng, D. Kong, “Automatic 3D liver location and segmentation via
convolutional neural networks and graph cut,” 2016, doi:10.1007/s11548-016-1467-3.
[31] R. Vivanti, A. Ephrat, L. Joskowicz, O. A, J. Sosna, “Automatic liver tumor segmentation in follow-
up CT studies using Convolutional Neural Networks The Rachel and Selim Benin School of
Computer Science and Department of Radiology ,” c, 2015.
[32] Q. Dou, L. Yu, H. Chen, Y. Jin, X. Yang, J. Qin, P. Heng, “3D deeply supervised network for
automated segmentation of volumetric medical images,” Medical Image Analysis, 41, 40–54, 2017,
doi:10.1016/j.media.2017.05.001.
[33] L. Massoptier, S. Casciaro, “Fully automatic liver segmentation through graph-cut technique,”
Annual International Conference of the IEEE Engineering in Medicine and Biology - Proceedings,
5243–5246, 2007, doi:10.1109/IEMBS.2007.4353524.
[34] L. Massoptier, S. Casciaro, “A new fully automatic and robust algorithm for fast segmentation of
liver tissue and tumors from CT scans,” 1658–1665, 2008, doi:10.1007/s00330-008-0924-y.
[35] A. Gotra, L. Sivakumaran, G. Chartrand, K.-N. Vu, F. Vandenbroucke-Menu, C. Kauffmann, S.
Kadoury, B. Gallix, J.A. de Guise, A. Tang, “Liver segmentation: indications, techniques and future
directions,” Insights into Imaging, 8(4), 377–392, 2017, doi:10.1007/s13244-017-0558-1.
[36] Mubashir Ahmad, Yuan Ding, Syed Furqan Qadri, Jian Yang, "Convolutional-neural-network-based
feature extraction for liver segmentation from CT images," Proc. SPIE 11179, Eleventh International
Conference on Digital Image Processing (ICDIP 2019), 1117934 (14 August 2019);
https://doi.org/10.1117/12.2540175
[37] Weiwei Wu, Zhuhuang Zhou, Shuicai Wu, Yanhua Zhang, "Automatic Liver Segmentation on
Volumetric CT Images Using Supervoxel-Based Graph Cuts", Computational and Mathematical
Methods in Medicine, vol. 2016, Article ID 9093721, 14 pages, 2016.
https://doi.org/10.1155/2016/9093721
[38] G. Litjens, T. Kooi, B.E. Bejnordi, A.A.A. Setio, F. Ciompi, M. Ghafoorian, J.A.W.M. van der Laak,
B. van Ginneken, C.I. Sánchez, “A Survey on Deep Learning in Medical Image Analysis,”
(February), 2017, doi:10.1016/j.media.2017.07.005.
[39] S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing
internal covariate shift,” 32nd Int. Conf. Mach. Learn. ICML 2015, vol. 1, pp. 448–456, 2015.

Eye Gaze Estimation Invisible and IR Spectrum for Driver Monitoring System

More Related Content

Similar to Eye Gaze Estimation Invisible and IR Spectrum for Driver Monitoring System (20)

Recently uploaded (20)

Eye Gaze Estimation Invisible and IR Spectrum for Driver Monitoring System