IRJET - Deep Learning based Bone Tumor Detection with Real Time Datasets

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 03 | Mar 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2391
DEEP LEARNING BASED BONE TUMOR DETECTION WITH REAL TIME
DATASETS
Santhanalakshmi .S.T 1, Abinaya.R 2, Affina sel. T.V 2, Dimple .P 2
1Assistant Professor(Grade 1), Department of Computer Science and Engineering, Panimalar Engineering College,
Chennai
2UG Scholar, Department of Computer Science and Engineering, Panimalar Engineering College, Chennai
----------------------------------------------------------------------***---------------------------------------------------------------------
Abstract-Digital image processing is the use of a digital
computer to process digital images through an algorithm.
Image Processing can be used to improve the pictorial
information for human interpretation. Since the cancerous
bone tumor is malignant which will destroy the cortex and
spread to other tissues of the body. Hence it is significant
to detect the bone tumor in earlier stage with more
accuracy. Medical Imaging has achieved a benchmark in
tumor detection by resolving various complexities. There
are different techniques in medical imaging like X-ray,
CT(Computerized Tomography), MRI(Magnetic Resonance
Imaging),Ultra sound. Among these various techniques
MRI gives with high accuracy rate and also MRI delivers
the best images as it has higher resolution. In this paper
the tumor detection has been proposed using deep
learning.
Keywords: Medical Imaging, Magnetic Resonance
Imaging, Recurrent Neural Network, Long Short Term
Memory.
1. INTRODUCTION
Bone tumor develops in skeletal system which destroys
tissues and spread to other parts of the body. There are
two types of bone cancer namely Primary and Secondary
bone cancer. Primary bone tumor either be benign (non-
cancerous) or malignant (cancerous).Non-cancerous
tumor do not spread beyond their original site whereas
cancerous tumors are more aggressive and have a higher
risk on growing and spreading. Secondary bone tumor
spreads to the bone from elsewhere in the body. Most
cancers can be spread to the bones of the body
.Moreover people with breast and prostate cancer have
high risk of developing secondary bone cancer. These are
the most common bone cancers in adults. Secondary
bone tumor is also known as bone metastases. For this
image processing plays a significant role in detecting
bone tumor. In the previous paper, to represent the local
structure of 3D patches in the vicinity of particular
location across the entire collection, Gaussian Mixture
Model(GMM) is used. Since in many clinical datasets the
thickness is unknown or varies by site, the existing
system could not explicitly model slice thickness.
In this paper, we propose bone tumor detection method
using Recurrent Neural Network(RNN) algorithm . In
medical examination data there often exist missing parts
due to various human factors because human subjects
occassionally miss their annual examinations. Thus
missing information need to get imputed for accurate
prediction of medical examination data. In our proposed
method, the trained RNNs are used both for missing data
imputation and target data prediction.
2. LITERATURE REVIEW
Haruna Watanabe et al [1] has proposed an Bone
Metastatic Tumor Detection Based on AnoGAN Using CT
Images. In this paper, Generative Adversarial
Network(GAN) based anomaly detection is done. The
proposed method uses only non-metastatic bone tumor
images and learns the distribution of normal images
based on adversarial learning. They define the anomaly
scores by comparing a test image with a generated image
.Metastatic tumor images can be automatically detected
based on calculated anomaly scores, but Still missing-
features problem exists and Computational analysis are
impractical .
Akash Pandey et al [3] has done a paper on A Survey
Paper on Calcaneus Bone Tumor Detection Using
different Improved Canny Edge Detector. In this paper
Computer Aided Diagnosing(CAD) is used to analyze
Computed Tomography images..And canny edge detector
is used for edge detection in image. But canny edge
detector has some limitations like it is not able to
distinguish edges occurring around objects but still it is
beneficial than other traditional edge detector method
since it improves the performance of image analysis
algorithms.
Eftekhar Hossain et al [4] has proposed an Comparative
Evaluation of Segmentation Algorithms for Tumor Cells
Detection from Bone MR Scan Imagery. This study
proposed an object labeling algorithm for the
segmentation of bone tumor from magnetic resonance
images (MRI) and also provides a comparative analysis
of the existing bone cancer segmentation methods . This
segmentation algorithms with the proposed one has
been compared on the basis of quantitative methods like
the dice similarity coefficient (DSC) and the 8structural
similarity index measurement (SSIM).Since it is
comparatively easier and provide greater noise

immunity over the edge detection method, the bone
images contain granny portions of tissues and low
volume tumor which make problems of over or under
segmentation
Eftekhar Hossain[2] has proposed an Detection &
Classification of Tumor Cells from Bone MR Imagery
Using Connected Component Analysis & Neural
Network. The bone tumor can be detected by using
connected component labeling algorithm . In this work,
for classifying the bone tumor Artificial Neural Network
(ANN) is used. Here bone MR images of previously
verified patients are collected and the texture features of
this images are used for the training and testing of the
neural network. Although ADF conserves the main edges
of the objects by removing the high- frequency noise and
thus, enhance the low volume region of the images. The
tumor portion will become shaded and thus difficult to
find out.
Krupali D et al [6] has done a paper on Integrated
Approach for Bone Tumor Detection from MRI Scan
Imagery. The paper proposes an approach of detecting
Enchondroma bone tumor from MRI images by using
image processing, segmentation clustering techniques,
i.e .K-means integrated with Fuzzy C-means clustering.
Detected manually by doctors, but because of noise and
low quality images of infected body parts, the tumor is
not detected easily and is time consuming.
NGOC-HUYNH HO et al [5] has proposed on
Regenarative Semi-Supervised Bidirectional W-Network-
Based Knee Bone Tumor Classification on Radiographs
Guided by Three-Region Bone Segmentation. This paper
develops and evaluate a new deep learning architecture,
namely regenerative semi-supervised bidirectional W-
network (RSS-BW), to predict the tumor state of theknee
bone from radiographic images. First, we constructed an
autoencoder model, called Bidirectional W-network
(BW), for segmenting three-region (i.e., femur, tibia, and
fibula) of knee bone. Using these regions as input data,
RSS-BW architecture consisting of the autoencoding
model for regenerating the bone structures, the
backbone model for extracting features with pretrained
Image Net, and the predicting model for knee bone
tumor classification are established.
3. PROPOSED SYSTEM
In our proposed system , in order to predict medical
examination data with missing parts Recurrent Neural
Network (RNN) is used. Among various types of RNNs.
we use simple recurrent network (SRN) and long short-
term memory (LSTM) in order to predict the missing
information along with the future medical examination
data, since these algorithm will show good performance
in many relevant applications. Basic RNNs are a network
of neuron-like nodes organized into successive layers. In
this algorithm each node in a given layer is connected
directly (one-way connection) to every other node in the
next successive layer. Due to this functionality the output
from the previous step are fed as input to the current
step. The important feature of Recurrent Neural Network
is Hidden state which remembers some previous
information about the sequence.
Fig-1:Implementation of
RecurrenNeuralNetworks(RNN)
Fig-2:Long Short Term Memory(LSTM) Networks
Long short-term memory (LSTM) is an artificial
recurrent neural network (RNN) capable of learning
long-term dependencies used in the field of deep
learning .Their default behaviour is to remember
information for long periods of time and it not only
process single data points (such as images), but also
entire sequences of data (such as speech or video).
LSTMs were developed to deal with the exploding and
vanishing gradient problems that can be encountered
when training traditional RNNs. The purpose of using
RNN and LSTM in our is to impute the missing data in
order to handle data with missing parts without extra
training data composed of missing
examples. In our proposed method, the
trained RNNs are used both for missing data imputation
and target data prediction .In the proposed system, when
there are no missing data, the RNN is processed
normally;
when there appears missing data, the output of the RNN
in the previous time step is used as
the input of the current time step. With such missing
data imputation method, the target data
with missing parts may be predicted by our proposed
RNNs.

Fig -3: System Architecture
3.1 MRI
Magnetic resonance imaging (MRI) scan uses radio
waves to look at organs and structures inside the body.
To diagnose various condition Health care professionals
use MRI scans from torn ligaments to tumors. There are
no risk or side effects of an MRI scan. The benefits of an
MRI relate to its precise accuracy in detecting structural
abnormalities of the body. Hence in our proposed system
we are using MRI images as datasets.
3.2 Image segmentation:
Multiple segments are formed using partition process
which is said to be image segmentation. The goal of
segmentation is that the resultant formed is something
meaningful to examine. The image has more pixel and
each pixel is assigned by a label. It shares some common
characteristics like same label etc... so segmentation
means contours extracted from an image.
3.3 Feature extraction:
Feature extraction reduces initial set of raw data into
more manageable groups for processing. Image
processing is a vast area to work with in which
behaviour of the image has been clearly extracted
through feature extraction techniques where some of
feature extraction technique are Histogram of oriented
gradients (HOG), Speeded-up robust features (SURF),
Local binary patterns (LBP), Haar wavelets, Color
histograms. Once the behaviour of the image can be
drawn then it is feasible to find the tumor in bones.
3.4 Tumor detection
Once the image features are extracted the detection of
bone tumor can be easily done since we are using MRI
images .Because MRI images gives more accurate
detection of tumor cells and we are using this image in
our system.
3.5 Tumor identification
The identification of tumor can be done by calculating
each pixels in the image and compare it with the trained
datasets. Once the comparison has been done it pin
points the particular point of affected area which gives
an accurate answer for identifying the bone tumor.
3.6 Diagnosis suggestion
Finally the diagnosis suggestion has been given to the
patient by analysing the MRI images which is the test
datasets that gives more accurate point of attack.
4. PERFORMANCE ANALYSIS
In our proposed system, we choose RNN for increasing
the accuracy of the bone tumor detection. In this user
just need to select the images and the rest of the process
can be handled by the algorithm. Since the missing parts
can be imputed in Recurrent Neural Network algorithm,
the accuracy can be increased when compare with the
Convolutional Neural Networks. The system result is in
the form of suggestion which increases the reliability of
the concept.
Fig-4: Vitamin D deficiency points

Fig -5: Vitamin D deficieny points[Sample between RNN and human]
5. CONCLUSION
We propose a bone tumor detection method by using
Recurrent Neural Network algorithm. Among several
types of RNNs, we choose Simple Recurrent Network and
Long Short Term Memory in order to impute the missing
parts in medical examination data .By using these
algorithm we can increase the accuracy of detecting the
bone tumor when compare with the existing system.
Thus our proposed system provides a different way for
detecting the bone tumor with high accuracy.
6. REFERENCES
[1] H. Watanabe, R. Togo, T. Ogawa and M. Haseyama,
"Bone Metastatic Tumor Detection based on AnoGAN
Using CT Images," 2019 IEEE 1st Global Conference on
Life Sciences and Technologies (LifeTech), Osaka, Japan,
2019, pp. 235-236.
[2] E. Hossain and M. A. Rahaman, "Detection &
Classification of Tumor Cells from Bone MR Imagery
Using Connected Component Analysis & Neural
Network," 2018 International Conference on
Advancement in Electrical and Electronic Engineering
(ICAEEE), Gazipur, Bangladesh, 2018, pp. 1-4.
[3] A. Pandey and S. K. Shrivastava, "A Survey Paper on
Calcaneus Bone Tumor Detection Using different
Improved Canny Edge Detector," 2018 ieee international
conference on system, computation, automation and
networking (icscan), Pondicherry, 2018, pp. 1-5.
doi: 10.1109/ICSCAN.2018.8541194
[4] E. Hossain and M. A. Rahaman, "Comparative
Evaluation of Segmentation Algorithms for Tumor Cells
Detection from Bone MR Scan Imagery," 2018
International Conference on Innovations in Science,
Engineering and Technology (ICISET), Chittagong,
Bangladesh, 2018, pp. 361-366.
[5] N. Ho, H. Yang, S. Kim, S. T. Jung and S. Joo,
"Regenerative Semi-Supervised Bidirectional W-
Network-Based Knee Bone Tumor Classification on
Radiographs Guided by Three-Region Bone
Segmentation," in IEEE Access, vol. 7, pp. 154277-
154289, 2019.
[6] K. D. Mistry and B. J. Talati, "Integrated approach for
bone tumor detection from MRI scan imagery," 2016
International Conference on Signal and Information
Processing (IConSIP), Vishnupuri, 2016, pp. 1-5.

IRJET - Deep Learning based Bone Tumor Detection with Real Time Datasets

More Related Content

What's hot (20)

Similar to IRJET - Deep Learning based Bone Tumor Detection with Real Time Datasets (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET - Deep Learning based Bone Tumor Detection with Real Time Datasets