Bleeding Frame

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.2015.2399502, IEEE Journal of Biomedical and Health Informatics
Bleeding Frame and Region Detection in the

Wireless Capsule Endoscopy Video
Yixuan Yuan, Baopu Li, Max Q.-H. Meng, Fellow, IEEE
the small intestines [2, 3]. The advancement of miniature

AbstractWireless capsule endoscopy (WCE) enables technology promoted the emergence of wireless capsule
non-invasive and painless direct visual inspection of a patients endoscopy (WCE) [4], which is a revolutionary imaging device
whole digestive tract, but at the price of long time reviewing large that provides direct, noninvasive visualization of the small
amount of images by clinicians. Thus an automatic
bowel. The WCE was first introduced by the Given Imaging in
computer-aided technique to reduce the burden of physicians is
highly demanded. In this paper, we propose a novel color feature 2000 and granted approval by U.S. Food and Drug
extraction method to discriminate the bleeding frames from the Administration in 2001. It is a capsule shaped device with a
normal ones, with further localization of the bleeding regions. Our dimension of 26mm in length by 11mm in diameter, which
proposal is based on a twofold system. First, we make full use of consists of an optical dome, an illuminator, an imaging sensor,
the color information of WCE images and utilize K-means a battery, and a RF transmitter. In the examination procedure,
clustering method on the pixel represented images to obtain the
the WCE is swallowed by the patient and then pushed by
cluster centers, with which we characterize WCE images as words
based color histograms. Then we judge the status of a WCE frame peristalsis to slowly travel through the small intestine.
by applying support vector machine (SVM) and K nearest Equipped with built-in lighting and camera, a WCE takes
neighbor (KNN) methods. Comprehensive experimental results pictures of the entire GI tract with 2-4 images per second for
reveal that the best classification performance is obtained with about 8 hours before its battery exhausted. Finally these images
YCbCr color space, cluster number 80 and the SVM. The are transmitted wirelessly to a data-recording device for doctors
achieved classification performance reaches 95.75% in accuracy,
to examine the images off-line later to make diagnostic
0.9771 for AUC, validating that the proposed scheme provides an
exciting performance for bleeding classification. Secondly, we decisions [5].
propose a two-stage saliency map extraction method to highlight Although the WCE has shown significant advantages, there
bleeding regions where the first stage saliency map is created by is still challenges associated with this technology. One problem
means of different color channels mixer and the second stage is that the WCE produces about 55,000 images for each
saliency map is obtained from the visual contrast. Followed by an patients examination, which makes it hard for physicians to go
appropriate fusion strategy and threshold, we localize the
through all these images frame by frame to locate the bleedings
bleeding areas. Quantitative as well as qualitative results show
that our methods could differentiate the bleeding areas from [6] in the GI tract. Furthermore, WCE images capturing the
neighborhoods correctly abnormalities of the GI tract occupy only a small percentage of
the complete WCE images collected [7]. In addition, although
Index Terms Bleeding classification and region detection, the software developed by the Given Imaging, named the red
words based color histograms, wireless capsule endoscopy. blood identification system (RBIS), can detect bleeding frames
automatically, it achieves low sensitivity (21.5%) and
I. INTRODUCTION specificity (41.6%) [8]. All these problems motivate us to turn
B LEEDING in the gastrointestinal (GI) tract result from a

number of etiologies, including vascular lesions, vascular
tumors, ulcers and inflammatory lesions [1]. The general
to computer-aided systems.
In the paper [9], the authors put forward a new method for
rapid bleeding detection. They grouped pixels through the
approach to diagnose the bleedings is to directly view the GI super-pixel segmentation procedure and used the red color ratio
tract by different manners. However, the traditional imaging in the RGB color space to represent the features of these
techniques such as push enteroscopy, sonde enteroscopy are not super-pixels. The authors in [10] utilized six color features
only painful and invasive, but also technically difficult to reach (mean and variance of H, S and V value) in the HSI color space
to discriminate between bleeding and normal status. Lv et al.
This project is partially supported by RGC GRF #CUHK415613 awarded to [11] introduced a new descriptor, that is, pyramid of hue
Max Q.-H. Meng and National Natural Science Foundation of China histograms (PHH) to characterize the bleeding WCE images,
(61305099). incorporating color and spatial information by combining
Yixuan Yuan is with the Department of Electronic Engineering, The
Chinese University of Hong Kong, N.T., Hong Kong SAR, China. Baopu Li is
illumination invariant color histograms and the spatial
with the Department of Biomedical Engineering in Shenzhen University, pyramids method. Although these methods can detect the
Shenzhen, China (corresponding author, email: bpli@szu.edu.cn). Max Q.-H. bleeding frames from the normal ones in some degree, majority
Meng is with the Department of Electronic Engineering, The Chinese
University of Hong Kong, N.T., Hong Kong SAR, China co-corresponding
of them extract the complete color features from a WCE image,
author; e-mail: qhmeng@ee.cuhk.edu.hk). ignoring the specific color range of WCE images.
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
Fig. 1 The workflow of the proposed method.
In this paper, we propose words based color histogram for experimental results validating the proposed intelligent system
bleeding detection in WCE images. Our method is an extension are presented and discussed in Section V. Finally, we draw
of Bag of Words method [12, 13]. In order to make the most of some conclusions and discussions in Section VI.
the color information of the bleeding images, we calculate the
color words by applying K-means clustering procedure to the II. METHOD OVERVIEW
pixel represented WCE images in the specific color space. Then The flowchart of the proposed method is depicted in Fig. 1
each WCE image is characterized as histogram of the cluster and it comprises two major steps. In the first step of bleeding
centers (named words based color histogram) to represent the frame classification, we propose words based color histograms
feature vector. Finally, support vector machine (SVM) [14] and to represent WCE images. Then we focus on the bleeding area
K nearest neighbor (KNN) [15] are utilized as classifiers to localization. We calculate the first stage saliency map based on
detect bleeding frames. In the second stage, we focus on the observation of the bleeding areas in the different color
localization of the bleeding areas in the bleeding frames. Since spaces and the second stage saliency map based on the natural
the components of various color spaces possess different property of bleeding areas in the RGB color space. Finally we
information, we inspect the bleeding images under different localize the bleeding areas through the fusion strategy of the
color spaces like RGB, HSI/HSV, CMYK, CIELAB, YUV, saliency maps and an appropriate threshold.
XYZ [16] and select the components that highlight the bleeding
areas. Then we create the first stage saliency map by combing III. BLEEDING FRAME DETECTION
these components together to strengthen the suspicious regions.
In addition, natural saliency lies in the visual contrast [17, 18]. A. ROI Extraction
Thus we extract the second stage saliency map from the prior sWCE images are often obscured by the large black
that if the color information of the region shows large similarity background and obvious bounders as shown in Fig. 2 (a), (b),
to the red color, then this region should possess high saliency thus the image features extracted from the entire image will
value. Finally applying an appropriate fusion strategy of these reflect the visual contamination presented in the image. To
two conspicuity maps and automatic threshold, we are able to address this factor, we outline the maximum square inscribed in
localize the bleeding area. the circular image as region of interest (ROI) without loss of the
Our main contributions can be summarized in the following major image information. The size of the obtained ROI is
two aspects. 180180 from the original image with the size of 256256. It
can be found that the extracted ROIs are satisfactory since they
1) We propose the words based color histogram to represent
demonstrate the major image features and provide a good
the images. That is, we make use of middle level features rather
characterization and description of the WCE images. The ROI
than low level features that are generally used in the literatures images replace the original ones for the following processing,
[7, 9-11]. Moreover, we study carefully the influence of the making the feature extraction procedure much easier.
color spaces, cluster centers and different classification
methods in terms of the classification performance.
2) We propose a two-stage saliency extraction method to
localize the bleeding areas in WCE images. Since these
two-stage saliency maps highlight the bleeding regions and
separate bleeding mucosa from the uninformative parts, we
could obtain the bleeding area candidates successfully.
The remainder of this paper is organized as follows. Section
II outlines the proposed method. Section III introduces the
Fig. 2 Illustration of ROI extraction.
bleeding frame classification method while Section IV
discusses the localization of the bleeding areas. The
B. Color Feature Extraction

Since clinicians discriminate bleeding frames from the whole
WCE images mainly based on the color information, thus
choosing a suitable color space is crucial, which converts colors
sensed by human eyes into given values such that computer
programs can use the corresponding color information. It is not
obvious in advance which color space is the most useful one to
expose the abnormality of bleeding. Thus we investigate the
proposed feature extraction methods in the commonly used
color spaces including RGB, HSV, YCbCr and LAB [1920]
and make a best choice through the experiment results.
After choosing the reference color space, we extract suitable
color features to describe bleeding frames in the endoscopic
images. Actually many existing approaches on feature
extraction focus on histograms in a complete color range and
statistic characteristics of the histograms. However, the WCE
videos may not include some colors like blue or violet [19]. Fig.3 Illustration of word based color histogram extraction method.
Furthermore, most of the observed colors in WCE images are
C. Bleeding Classification Method and Criteria
concentrated in a small region of the color space [20]. To
address these issues, we proposed a novel color feature to To verify the performance of the proposed feature, we
characterize the WCE images. deploy two common classification methods: SVM and KNN to
In our proposed color feature extraction method, as shown in evaluate their powers in differentiating normal images and
Fig.3, in order to obtain the specific color range of the WCE bleeding images. SVM [14] is a supervised machine learning
images, we randomly select 10% bleeding images and 10% method based on the foundation of statistical learning and we
normal images from the datasets and calculate the referred to the work of Chang and Lin in [22] to carry out the
corresponding cluster centers independently by inputting the experiments. In our method, the radial basis function is used as
pixel represented image vectors in the color space to the SVM kernel.
K-means clustering procedure [21]. This procedure utilizes the KNN is a kind of simple and intuitive method widely used in
color information of the WCE frames and reduces the pattern recognition and data mining. In general, we calculate
dimension of the color feature. The concatenated cluster centers the Euclidean distance from the new test image to all the
K from the bleeding dataset and the normal WCE dataset serve training samples in the feature space and classify this image
as a vocabulary of visual words. Then we map based on the K closest neighbors from the training set. The
three-dimensional color data of each point in a WCE image to label that comprises the majority of the K nearest training
the nearest visual words and calculate the number of each samples is assigned to the test sample. In this experiment, we
set K to 10 based on the optimal performance to conduct the
visual word, yielding a histogram ( w, d ) , where wi denotes
experiment.
the i th visual word in the K-size color clusters and d i counts The bleeding frame classification performance in the WCE
images is measured by accuracy, specificity and sensitivity,
the frequency of occurrence of it. Utilizing this method, we
which are widely used to evaluate classification performance.
characterize the WCE images as the words based color
The sensitivity shows capability of detecting bleeding images
histograms. An important decision in the construction of the
while specificity means the ability to avoid false detection.
new feature is the selection of the vocabulary size K. To
Accuracy is used to assess the overall performance of the
evaluate how the vocabulary size influences the classification
algorithm, which reflects sensitivity and specificity in relation
performance with our approach, we gradually increase K from
to each other.
10 to 100 and evaluated the classification results under each
setting.
IV. BLEEDING LOCATION DETECTION
As discussed above, the proposed color features are based on
the histogram, thus it preserves the robustness to the rotation A. First Stage Saliency
and translation of the image contents. In addition, the words After obtaining the bleeding frames, we focus on the
based color histograms make the best of the range of the color bleeding area localization in this part. Since color is the main
information in the WCE images and represent the color cue used to distinguish the bleeding areas from the normal
distribution of the images in a middle level with smaller mucosa and it is not obvious in advance which color component
dimensions than the traditional color features. Therefore, the contains the most useful information to expose the abnormality
proposed novel feature could provide an accurate description of of bleeding, therefore, we inspect the bleeding images under
the WCE images effectively. different color components of various color spaces such as
RGB, CIELAB, CIEXYZ, YCbCr, CMYK, HSV and HSI. As
shown in Fig.4, interestingly, we have observed that the second
component of the transformed WCE images in the CIELAB [23]
and CMYK [24] color spaces highlights the bleeding regions
and separates bleeding mucosa tissues from the uninformative VB2

parts. S B ( x, y ) exp( ) (4)
B
where VR , VG and VB are the corresponding values in the RGB
space. The parameters R G and B are chosen as
R =0.32, G B 0.45 to adjust the three color saliency
map to [0, 1]. It can be found that in the formula (2)-(4), the
region with larger VR , smaller VG and VB will have higher S R ,
SG and S B .
Finally the second stage saliency map S stage 2 for an image I
could be formulated as the combination of the three channel
saliency maps:
1
Sstage2 ( x, y) ( S R SG S B ) (5)
3
C. Fusion Scheme of Two Saliency Maps
The above two saliency maps localize the bleeding areas in
the WCE images from different aspects, one is from the color
space transformation and the other is from the view of the
Fig.4 Original frames with bleeding regions in (a), ROIs in (b), second channel clinician. Thus we could derive our proposed saliency map by
of CIELAB color space in (c), second channel of CMYK color space in (d). fusing them together by the following formula:
S final w1 * Sstage1 w2 * Sstage 2 (6)
In order to emphasize the bleeding mucosa, a saliency map
is created by assigning different weights to the aforementioned where S final is the corresponding fused saliency map and
two planes that highlight the bleeding areas with the following
equation: w1 w2 1 . We adjust w1 from 0.1 to 1 with 0.1 increments
Sstage1 ( x, y) * A( x, y) * M ( x, y) (1) each time to evaluate the different weights to the final bleeding
localization. Once these two stage saliency maps are fused
where A( x, y ) represents the A channel of the CIELAB together, the bleeding regions are likely to be strengthened.
color space [23] while M ( x, y) is the M channel of the Finally after applying a thresholding algorithm, we could
CMYK color space [24]. , are predefined constants as further localize the true bleeding regions, the saliency map will
be further processed into a binary mask image.
0.5 in our experiments. In addition, since the range
D. Criteria for bleeding localization
of the A plane in CIELAB space and the M plane in the
CMYK plane is different, we normalize these two color planes To quantitatively assess the localization performance, a
before applying the formula (1). pixel-based comparison between the localized bleeding regions
and the ground truth labeled by the clinicians is performed. The
B. Second Stage Saliency three metrics used in this experiment are Precision, the false
When clinicians view the WCE images, they may be positive ratio (FPR) and the false negative ratio (FNR) [25, 26],
attracted by the bleeding areas based on red color in RGB space which are calculated according to the following formulas:
instead of the transformations in the different color spaces.
TP (8)
Therefore, we define the second salient regions as those regions Precision
TP FP
with large similarity to the red color values, which means that if
a pixel has a greater R value and smaller G, B values, it would FP (9)
FPR
seem reddish and should be assigned higher saliency value. FP TN
Hence we derive the second stage saliency map based on the
FN (10)
following three steps. Firstly we utilize a 5*5 Gaussian filter on FNR
the original image to eliminate fine texture details as well as FN TP
noise and coding artifacts. Then the saliency maps for R, G and where true positive (TP) cases are the bleeding areas that are
B color channels are calculated by correctly labeled as the bleedings while false positive (FP) are
V2 the ones incorrectly labeled as the bleedings. False negative
S R ( x, y) 1 exp( R ) (2)
R (FN) represent the regions which are not labeled as the
bleedings but should have been.
VG2
SG ( x, y ) exp( ) (3)
G
V. EXPERIMENT RESULTS
A. Image acquisition and experimental setup
In our experiments, we conducted experiments on 2400
WCE images that consist of 400 bleeding frames and 2000
normal frames. These images were extracted from 10 different
patients videos and manually annotated by gastroenterologists.
The effective resolution of these images is 256*256.
B. Experiment Results for Bleeding Frames Detection
1) Parameters Selection
The first experiment was designed to evaluate the
discriminative power of the proposed feature and choose the
best parameters in the classification of the bleeding WCE
images. Fig. 6. Feature analysis. (a) Normal images and the corresponding feature
The most important two parameters in the construction of the histogram. (b) Bleeding images and the corresponding feature histogram.
word based color histogram are the selection of the vocabulary
size K and the color space. To evaluate how these two 3) Comparison to the other methods
parameters influence the classification performance, we To further evaluate the performance of the proposed new
conduct the coupled experiments. We tune K from 10 to 100 feature, we compared it with the six state-of-the-art color
with 10 increments in the experiment and the results for the features. The first two comparison methods directly used the
different classification methods SVM and KNN in different HSV histogram and YCbCr histogram to describe images.
color spaces are showed in the Fig. 5. To be mentioned here, we These histograms are produced by dividing the colors in the
image into specific number of bins and counting the number of
apply a 10-fold cross validation method to validate our
image pixels in each bin. For the HSV histogram, it quantizes
proposed method [27].
the whole color space into 72 main colors (dividing H into eight
We first analyzed the classification performance with
parts, S and V into three parts) while for the YCbCr histogram,
different classification algorithms. It can be found that the 125 bins are used to represent image features (dividing Y, Cb,
SVM method (Fig. 5 (a)-(c)) is much more suitable than the Cr into 5 parts). The method in [10] and [11] has already been
KNN method (Fig. 5 (d)-(f)) for bleeding classification in the mentioned in the introduction part and these two papers are the
WCE images because the SVM method shows relative higher latest related papers focused on the bleeding detection. We also
accuracy, specificity and sensitivity in all of the color spaces. compared the proposed method with latest feature coding
We then studied the classification performance with the SVM method: sparse coding [28] and Locality-constrained linear
classifier. As illustrated in Fig. 5 (a)-(c), there are some coding (LLC) [29]. In these two methods, after obtaining the
differences among the four color spaces. Overall the YCbCr cluster centers, instead of represent image by vector
color space exhibits the best classification performance while quantization, they encode each point in the image by the sparse
the HSV color space is not suitable for bleeding detection with coding and LLC method to obtain the final histogram. We
lowest accuracy compared with other color spaces. implemented these algorithms on our datasets and the
In addition, the cluster center size does affect the corresponding comparison results are showed in Table I.
classification performance. In the YcbCr color space, the TABLE I
classification performance increases first and then decreases PERFORMANCE COMPARISON OF DIFFERENT FEATURES (%)
when the cluster size increases. The best classification result is HSV YCbCr [10] [11] [28] [29] Our
histo- histo- me-
achieved with cluster centers 80 and the corresponding gram gram thod
accuracy and sensitivity reach 95.75%, 92.00%. The AUC of Acc. 92.37 86.25 90.00 83.00 91.63 76.63 95.75
the classification with the best parameter setting is 0.9771, Sen. 84.75 91.50 81.75 78.00 94.75 91.25 92.00
Spec. 93.90 85.20 91.65 84.00 91.00 62.00 96.50
validating the proposed feature has a good capability to classify
Time(s) 75.93 422.27 35.76 530.43 4171.5 28737 293.43
bleeding images.
2) Feature analysis It can be found in Table I, the proposed method shows better
To show the effectiveness of our features, we plotted two performance to those of the HSV histogram method with an
examples of the feature histogram with cluster number 80 for improvement 3.38% and 7.25% of in accuracy and sensitivity,
the normal images and bleeding images in the Fig. 6. The respectively, while to those of YCbCr histogram with an
cluster centers 1-40 are obtained by the bleeding images while improvement 9.5% and 0.5%. This result demonstrates that the
the cluster centers 41-80 are obtained from the normal images. words based color histogram is superior compared with the
We could find that the bleeding image and the normal image traditional color histograms to classify the bleeding frames. In
addition, our method gets higher accuracy, sensitivity and
show significant different words based color histograms.
specificity than the existed method [10] and [11] in the bleeding
frame classification task. Compared with the latest feature
coding method [28] and [29], our method not only shows better
accuracy and sensitivity, but also is much faster than sparse
Fig. 5 Classification performance with different color spaces and classification method. (a)-(c) the classification performance with SVM. (d)-(f) shows the
classification performance with KNN.
coding and LLC, which demonstrating our method is more 2) Qualitative Analysis
effective in clinical situation. We fixed w1 0.8 to illustrate our proposed saliency
C. Experiment Results for Bleeding Localization methods visually by the first example image given in Fig. 5. Fig.
1) Quantitative Analysis 9(a) is the saliency map generated through means of channel
After obtaining the bleeding frames, we focused on the mixer while Fig. 9(b) is obtained by prior knowledge that the
localization of the bleeding areas in this part. Our model region with red color should assign large saliency value. It is
calculates two stage saliency maps and fuses them together to obvious that these two saliency maps provide good ability to
detect the bleeding areas. Evaluation of the fusion strategy is an identify bleeding regions. After the fusion strategy, the final
important step in the assessment of the model performance. We saliency map is shown in Fig. 9(c). In this figure, the bleeding
mucosa appears brighter than normal mucosa region, hence it
tested w1 in the Eq. 6 for the values {0,0.2,0.4, 0.6,0.8,1} to
can be easily extracted by an appropriate threshold. Fig. 9(d) is
obtain the final saliency maps and then applied the Otsus [30] the binary bleeding mask image obtained through the Otsus
threshold method on the calculated saliency maps to obtain the threshold procedure.
bleeding areas.
TABLE II
PERFORMANCE COMPARISON OF DIFFERENT FOR w1 BLEEDING
LOCALIZATION (%)
W1 Precision FPR FNR W1 Precision FPR FNR
0 66.27 10.71 34.09 0.6 94.70 0.99 36.79

0.2 78.89 5.27 34.21 0.8 95.24 0.86 38.97
0.4 91.05 1.82 34.99 1 93.97 1.17 33.72
To evaluate the localization performance that was obtained

by using different weighting models, three criteria: Precision, Fig. 9. The procedure of localization for bleeding area. (a) First stage saliency
FPR, FNR were calculated and the corresponding results are map. (b) Second stage saliency map. (c) Fused saliency map. (d) The localized
showed in the Table II. The best bleeding localization result bleeding area.
with the precision of 95.24% was achieved with weight of 0.8 In Fig. 10, qualitative results for the bleeding area
on the first stage saliency map. This result is inspiring, showing localization based on the saliency maps are presented using
that our proposed saliency method could detect the bleeding three different WCE ROI images containing bleedings. The
area accurately. Moreover, when the weight of the saliency first column is the ROI images while the second column shows
region is set to 0 or 1, the best localization accuracy is not the bleeding saliency map, and the third column shows the
obtained. This result indicates that these two-stage saliency detected bleeding region. The final column represents the
calculation methods complement each other to provide useful ground truth for the bleeding area labeled by the clinicians. As
information for bleeding localization. Furthermore, the can be observed, we are able to differentiate the bleeding areas
algorithm achieved a good FPR of 0.86% with w1 0.8 . accurately by the proposed bleeding localization method.
Information and Automation (ICIA), 2010 IEEE International

Conference on, 2010, pp. 1746-1751.
[11] G. Lv, G. Yan, and Z. Wang, "Bleeding detection in wireless
capsule endoscopy images based on color invariants and spatial
pyramids using support vector machines," in Engineering in
Medicine and Biology Society, EMBC, 2011 Annual International
Conference of the IEEE, 2011, pp. 6643-6646.
[12] J. Sivic and A. Zisserman, "Video Google: A text retrieval approach
to object matching in videos," in Computer Vision, 2003.
Proceedings. Ninth IEEE International Conference on, 2003, pp.
1470-1477.
[13] G. Csurka, C. Dance, L. Fan, J. Willamowski, and C. Bray, "Visual
categorization with bags of keypoints," in Workshop on statistical
learning in computer vision, ECCV, 2004, pp. 1-2.
[14] C. Cortes and V. Vapnik, "Support-vector networks," Machine
learning, vol. 20, pp. 273-297, 1995.
[15] S. A. Dudani, "The distance-weighted k-nearest-neighbor rule,"
Systems, Man and Cybernetics, IEEE Transactions on, pp. 325-327,
1976.
[16] G. Sharma and H. J. Trussell, "Digital color imaging," Image
Processing, IEEE Transactions on, vol. 6, pp. 901-932, 1997.
[17] R. Achanta, F. Estrada, P. Wils, and S. Ssstrunk, "Salient region
Fig. 10. Three examples of localization for bleeding areas. (a) ROI images. (b) detection and segmentation," in Computer Vision Systems, ed:
Final saliency maps. (c) The localized bleeding areas. (d) Ground truth. Springer, 2008, pp. 66-75.
[18] S. Goferman, L. Zelnik-Manor, and A. Tal, "Context-aware
VI. CONCLUSIONS saliency detection," Pattern Analysis and Machine Intelligence,
IEEE Transactions on, vol. 34, pp. 1915-1926, 2012.
In this paper, we have proposed a novel method for bleeding [19] S. Segui, M. Drozdzal, F. Vilarino, C. Malagelada, F. Azpiroz, P.
frame detection and region localization in WCE images. The Radeva, et al., "Categorization and Segmentation of Intestinal
extensive experiments demonstrate that the best classification Content Frames for Wireless Capsule Endoscopy," Information
Technology in Biomedicine, IEEE Transactions on, vol. 16, pp.
performance could be obtained with SVM classifier, YCbCr 1341-1352, 2012.
color space and cluster number of 80. The proposed features [20] S. Hwang, J. Oh, J. Cox, S. J. Tang, and H. F. Tibbals, "Blood
could obtain accuracy 95.75%, sensitivity 92% and specificity detection in wireless capsule endoscopy using expectation
96.5%, and the corresponding AUC is 0.9771. In the second maximization clustering," in Medical Imaging, 2006, pp.
61441P-61441P-11.
step, we extracted two stage saliency maps to locate the final [21] T. Kanungo, D. M. Mount, N. S. Netanyahu, C. D. Piatko, R.
bleeding areas. The quantitative result indicates the best Silverman, and A. Y. Wu, "An efficient k-means clustering
bleeding localization performance could be obtained by the algorithm: Analysis and implementation," Pattern Analysis and
weight 0.8 for the first stage saliency map and 0.2 for the Machine Intelligence, IEEE Transactions on, vol. 24, pp. 881-892,
2002.
second stage saliency map. The corresponding localization [22] C.-C. Chang and C.-J. Lin, "LIBSVM: a library for support vector
precision archives 95.24%. In the future, we will find more machines," ACM Transactions on Intelligent Systems and
suitable features for the bleeding frames classification in order Technology (TIST), vol. 2, p. 27, 2011.
to decrease the mistakes in the localization tasks. [23] J. Schanda, Colorimetry: Understanding the CIE system: Wiley.
com, 2007.
[24] Y. Zhao, Z. Fan, and M. E. Hoover, "Frequency domain infrared
REFERENCES watermarking for printed CMYK image," in Image Processing
[1] D. O. Faigel and D. R. Cave, Capsule endoscopy: Saunders (ICIP), 2011 18th IEEE International Conference on, 2011, pp.
Elsevier, 2008. 2725-2728.
[2] M. Yu, "M2a (tm) capsule endoscopy: A breakthrough diagnostic [25] F. Conversano, E. Casciaro, R. Franchini, S. Casciaro, and A.
tool for small intestine imaging," Gastroenterology Nursing, vol. Lay-Ekuakille, "Fully automatic 3D segmentation measurements of
25, pp. 24-27, 2002. human liver vessels from contrast-enhanced CT," in Medical
[3] G. Gay, M. Delvaux, and J.-F. Rey, "The role of video capsule Measurements and Applications (MeMeA), 2014 IEEE
endoscopy in the diagnosis of digestive diseases: a review of current International Symposium on, 2014, pp. 1-5.
possibilities," Endoscopy, vol. 36, pp. 913-920, 2004. [26] F. Conversano, R. Franchini, C. Demitri, L. Massoptier, F.
[4] G. Iddan, G. Meron, A. Glukhovsky, and P. Swain, "Wireless Montagna, A. Maffezzoli, et al., "Hepatic vessel segmentation for
capsule endoscopy," Nature, vol. 405, p. 417, 2000. 3D planning of liver surgery: experimental evaluation of a new fully
[5] N. M. Lee and G. M. Eisen, "10 years of capsule endoscopy: an automatic algorithm," Academic radiology, vol. 18, pp. 461-470,
update," 2010. 2011.
[6] M. Pennazio, "Capsule endoscopy: Where are we after 6 years of [27] Q. Li and K. Doi, "Reduction of bias and variance for evaluation of
clinical use?," Digestive and Liver Disease, vol. 38, pp. 867-878, computer-aided diagnostic schemes," Medical physics, vol. 33, pp.
2006. 868-875, 2006.
[7] B. Li and M.-H. Meng, "Computer-aided detection of bleeding [28] J. Yang, K. Yu, Y. Gong, and T. Huang, "Linear spatial pyramid
regions for capsule endoscopy images," Biomedical Engineering, matching using sparse coding for image classification," in
IEEE Transactions on, vol. 56, pp. 1032-1039, 2009. Computer Vision and Pattern Recognition, 2009. CVPR 2009. IEEE
[8] R. Francis, "Sensitivity and specificity of the red blood Conference on, 2009, pp. 1794-1801.
identification (RBIS) in video capsule endoscopy," in The 3rd [29] J. Wang, J. Yang, K. Yu, F. Lv, T. Huang, and Y. Gong,
International Conference on Capsule Endoscopy, 2004. "Locality-constrained linear coding for image classification," in
[9] Y. Fu, W. Zhang, M. Mandal, and M.-H. Meng, "Computer-Aided Computer Vision and Pattern Recognition (CVPR), 2010 IEEE
Bleeding Detection in WCE Video," Biomedical and Health Conference on, 2010, pp. 3360-3367.
Informatics, IEEE Journal of, vol. 18, pp. 636-642, 2014. [30] N. Otsu, "A threshold selection method from gray-level
[10] L. Cui, C. Hu, Y. Zou, and M.-H. Meng, "Bleeding detection in histograms," Automatica, vol. 11, pp. 23-27, 1975.
wireless capsule endoscopy images by support vector classifier," in

Bleeding Frame

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bleeding Frame

Uploaded by

Copyright:

Available Formats

This article has been accepted for publication in a future issue of this journal, but has not been

Bleeding Frame and Region Detection in the

the small intestines [2, 3]. The advancement of miniature

B LEEDING in the gastrointestinal (GI) tract result from a

Fig. 1 The workflow of the proposed method.

B. Color Feature Extraction

and separates bleeding mucosa tissues from the uninformative VB2

0 66.27 10.71 34.09 0.6 94.70 0.99 36.79

To evaluate the localization performance that was obtained

Information and Automation (ICIA), 2010 IEEE International

You might also like