Volume 4, Issue 10, October – 2019 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Web Based Voice Controlled Advanced PACS to

Diagnose Lungs Cancer and Related Anomalies
Thusiyanth Ratnasingam Abiram Sayanthan Evanjeline Sajeetha Velummylum
Faculty of Computing Faculty of Computing Faculty of Computing
Sri Lanka Institute of Information Sri Lanka Institute of Information Sri Lanka Institute of Information
Technology Technology Technology
Malabe, Sri Lanka Malabe, Sri Lanka Malabe, Sri Lanka

K.Archchana Koliya Pulasinghe

Faculty of Computing Faculty of Computing
Sri Lanka Institute of Information Technology Sri Lanka Institute of Information Technology
Malabe, Sri Lanka Malabe, Sri Lanka

Abstract:- PACS (picture archiving and communication malignant or benign. If the patient has conform the cancer
system) is a medicinal imaging technology that is hospital requiring to get the MRI and CT, then analyzation
utilized basically in medicinal services associations to of images in diagnosing lung cancer, 3D multiplications of
safely store and dissect carefully transmit electronic CT and MRI outputs makes them profoundly exact what's
pictures. PACS a needed asset in contemporary more, explicit to every patient. These 3D prints give
hospitals, has demonstrated its key position in the specialists more data than what's conceivable in a 2D
department of radiology for archiving and collecting portrayal of a patient's life structures in an itemized
medical images, followed by its inclusion with the manner. Identifying the boundaries of lung cancer to
department of radiology. In this paper we have included ascertain the volume of it will be helpful in finding
work 3D displaying of DICOM pictures, calculation of accurate volume calculation with less errors.
segmented cancer part with fine calculations, voice
recognition for a program to get and translate Voice transcription programming is gradually
correspondence or to comprehend and complete spoken supplanting the medical job that a medical transcriber. By
directions, and forecast of malignancy utilizing moving endlessly from a transcriber, the specialist is set in
examinations of cancer symptoms. a position where they have more noteworthy command
over their patient records here we use actions that can be
Keywords:- MRI, DICOM, K-NN, SVM, FNA, PACS) done using voice controls rather than manual operation in
the system e.g.: Rotate the 3D model.
I. INTRODUCTION
II. LITERATURE SURVEY
In this modern world, Healthcare medical imaging
system plays and important and major role in critical factor Now if we take Srilanka due to growth of the
for the quality of diagnostic and treatments. Sri Lanka population an d The increase amount of diseases and in
holds a one of a kind position in South Asia as one of the busy life schedule having manual reports of patients and
first of the less created countries to give general wellbeing handling that is very difficult and it also can make way to
Even though many technologies arrived in medical field, many errors and misconceptions [1].Using the PAC system
due to the poor knowledge in medical technology it cannot generated MRI DICOM images can be modelled in to 3D .
do any impact or solve the problems. Sri Lanka needs more Immense success can achieve by the Doctor in surgeries
easy to use and simpler to handle system so as to use in and also calculation volume and can able to identify the
occupied calendar. Picture archiving and Communication severity of the condition by observing 3D model rather than
System (PACS) is the back-bone of the investigation about 2D DICOM Images. Visualizing the brain tumor and
medicine pictures as it is well adjusting a few gauges, for calculating its volume are features of the proposed system
example, DICOM. It gives efficient capacity, recovery, the [4].
executives, conveyance and introduction of medicinal
pictures. Electronic pictures and reports are transmitted There are some algorithms and methods used to make
carefully by means of PACS. Breas cancer prediction Jiminguo1, BenjaminC. M. Fungs,
Farikhund introduced the Leiden University Medical
Breast cancer is one of the most prevalent cancers Center choice tree algorithm with breast cancer information
among women in the world, accounting for the majority of sets [1]. The information sets have 574 patients at that
new cases of cancer and cancer deaths, according to global hospital who have had surgery. Thus, within three years of
statistics. The aim of this assessment is to determine which original diagnosis, they produce the recurrence of breast
characteristics are most useful in predicting cancer that is cancer through a Decision Tree Algorithm. The classifier

IJISRT19OCT2218 www.ijisrt.com 631

Volume 4, Issue 10, October – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
anticipated precision of 70 percent. The Classifier precisely III. METHODOLOGY
predicters the recurrences of the disease in 55 patient for
the autonomous Classifier of 65 patient. The classifier also The scheme has been created using codeIgniter,
divides patients into two based on the trait of their disease tensorflow. Jupyter's ML notebook. MATLAB was used
and their significance to early recovery. There is a big for image processing and wamp, xampp was used for
amount of articles on the application of Data Mining server and database.
methods for survivability assessment [2] .The most of the
articles on the application of Data Mining methods for A. Voice Recognition and system controlling
survivability assessment. Several studies have been Speech is an extremely appealing way to interact with
reported to have focus 0n the significance of breast cancer human computers. It’s free of hands. It only takes normal
method. These studies have implemented distinct hardware for use (a HighcQualitycMicrophone or
approaches to data mining and provided problems, NormalzMicrozphones) [6]. Fundamentals of Speech
Decision Tree, Artificial Neural Network, Genetic Recognition comes atzazverys simple bits rate. Recognin
algorithms, Support Vector Machines peoples speech, especially continuous speech without
loaded training for a .vocabulary. of enough complexity is
Some algorithms and techniques are used to model very hard. By using new processes, flow diagrams,
3D from 2D medical pictures. For small configuration algorithms, and methods we can process speech signals easy
computers, Multi-Planning rendering (MPR) does not and recognize speech-text. In this system it develops a
require too many calculations. Method Volume Render single line speech-to-text system [7].
(VR) is used to visualize the object's full volume.
Cloudines and color must be calculated at each Voxel by The following are the voice production elements that
projecting rays through the volume data [2]. are examined while implementations use distinct voice
related functionalities. They are Phonation (producing
Finding volume can helps in finding the severity of sound), Fluency, Intonation, Pitch variance, Voice
the cancer and the risk of the operation. Estimates of tumor (including aeromechanical components of respiration).
volume provide only anatomical information and not true Basic voice recognition components are the sound card,
functional information [3]. There are some current systems microphone, speaker profile, language mode
accessible for segmentation of the brain tumor. M. Karuna
and Ankita Joshi suggested an "Automatic detection and Utterances Grouping and analyzing the words,
assessment of brain tumor using MATLAB" system [4]. matching the words, converting to text. The speaker-
This scheme includes segmentation through the Neuro recognition scheme can be seen as functioning in four
Fuzzy Classifier, but many input pictures are needed to phases [8].
train an image to the neural network.  Analyzing.
 Features extractions.
Now if we take Sri Lanka due to increase of the  Modeling.n
population and the increase amount of diseases and in busy  Test.
life schedule having manual reports of patients and handling
that is very difficult and it also can make way to many
errors and misconceptions [5]. There are some software
packages for speech recognition that are accessible
commercially for speech recognition. IBM Via Voice 98
with Generals Medical Vocabulary’s , Dragons Systems
naturally-speaking Medicala aSuite, version 3 ; and General
Medicine Edition, L and H Voice-xpress[6 ]. Twelve
doctors completing their training with each software
package compare these software packages and dictated a
summary of medical advancement and discharge. The
measurements analyzed in the research [6] are medical
abbreviations, medical vocabulary and general English
vocabulary compared across package using a standardized,
strict scoring strategy. IBM software’s was discovered to
have the smallest general errors rating software’s relative to
other, consecutive generations of speech recognition. The Fig 1:- Voice recognition process
item is extremely costly for use in developing nations such
as Sri Lanka, although it is successive. Capturing the voice and removing unwanted
Creating medical reports using voice recognition is done as background noise. Converting to analog to digital format.
part of the PAC system with a minimum cost and as the hig The models are intended either for a particular speaker or
h purpose of saving physicians time. PAC system will be for an autonomous speaker. The manner in which the
the solution to the above problem. It is more convenient and speaker talks also plays a part in recognizing speech. Some
user friendly so that and medical staffs can learn it and make models, with a pause in between, can acknowledge either
use of it.

IJISRT19OCT2218 www.ijisrt.com 632

Volume 4, Issue 10, October – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
single. Testing data to test the model .then converting to the C. Calculate the Tumor Volume
API.
 Hidden Markov Random Field Model
B. Breast cancer prediction HMRF-EM method of MRI segmentation. This
In this research are aim is to classify tumors as method has been effectively connected to tissue
malignant (cancerous) or benign (non-cancerous) using segmentation. MRI images Segmentation by utilizing an
features from multiple cell pictures. To calculate alternate count and slice size of pictures. Hidden Markov
characteristics, a digitized image of a Fineg Needlea Random Field model is gotten from Hidden Markov Model
aspiration (FNA) of a breast mass is used. They portray [10] [11]. This is characterized as procedures created by a
features of the image's cell’s nucleus. Breast cancer Markov chain. State progression can't be watched
information from the kajal internet was gathered [9]. straightforwardly, yet can be seen after certain means [12].

There are 530 instances with 33 attributes in the 𝑃(𝑦|𝑥) = ∏ 𝑃(𝑦𝑖 |𝑥𝑖 )
collected dataset. The class allocation is framed as unlikely,
𝑖∈𝑠
likely and beneficial. Therefore, there are 32 independent In a HMRF model y is an observable random field & x
variables and 1 variable dependent. For the independent is a hidden random field.
variables and dependent variable, the nominal values are set.
The non-modifiable factors are the first 10 variables and the 𝑃(𝑥1, 𝑥2, 𝑥3,……., 𝑥𝑛, 𝑦1, 𝑦2,𝑦3,……., 𝑦𝑛, )
modifiable factors are the next 22 variables and the last 𝑛
variable is the danger of breast cancer. = 𝑃(𝑦1 )𝑃(𝑥1 |𝑦1) ∏ 𝑃(𝑦𝑘 |𝑦𝑘−1 ) 𝑃(𝑥𝑘 |𝑦𝑘 )
𝑘=2
 Logistic Regression Algorithm
It is the go to technique for issues with binary This equation is used to find the probability.
classification. It is applied in predictive modeling [15].
Logistic regression is called for the perform used at the core  Hmrf Using The Em Algorithm
of the strategy, the supplying perform. Logistic Regression
Algorithm has the highest accuracy value 0.98. 𝑥 ∗ = arg max{𝑝 (𝑦|𝑥, 𝜃)𝑃(𝑥)}
 K-NN Algorithm Calculation is used to choose the sample parameter θ
In pattern recognition, the K-Nearest Neighbor for every pixel and comprises of two stages. First
formula may be a non-parametric methodology used for incorporates the estimation of the absentminded data which
classification and regression. In each cases, the input is required to shape an all-out enlightening accumulation
consists of the K Nearest training examples within the and the second incorporates the development of the normal
feature house [6]. K-NN may be a sort of instance-based probability work for the absolute instructive gathering. We
learning’s-nn algorithm accuracy value 0.97. search for the naming x* which fulfills.
 SVM Algorithm
𝑃(𝑥) = 𝑍 −1 exp(−𝑈(𝑥) )
The objective of the supporting vector machine-
leering algorithm is to find a hyper plane in an N-
MRF can proportionately be portrayed by a Gibbs
dimensionalx space. The accuracy value is 0.97.
Dissemination.
 Decision tree algorithm
 Threshold Method
Decision-Trees-Classifier is an easy decisions learn
This is utilized to change over unique picture to
algorithms which only recognizes categorically data for
binary picture. When the lung picture darker than the
model building. The fundamental concept is to build a
background picture, [13] [14]
DecisionTree by using a top to down-greedyz-searching
through-training the data set to evaluate each attribute at
Gray level pixel < T equal black color.
each node. It-utilizes-statistical property known as gaining
Gray level pixel > T equal white color.
data to select which attribute to test at each node in the tree.
Information gaining measures how well a specified attribute
distinguishes-the-training-sample-according to  Voxel Images
Voxel is a solitary pixel that given by: (Length *
theirpclassification. The accuracy value is 0.95.
Width * Slice size). [15]

To improve the output, HMRF-EM algorithm wants

to marginally wide edges of the distinguished articles in the
MRI scan image and accomplished by utilizing Gaussian
blur filter algorithm. The G (x, y) is

1 −(𝑥2 +𝑦2 )
𝐺(𝑥, 𝑦) = 2
𝑒 2𝜎2
√2𝜋𝜎

IJISRT19OCT2218 www.ijisrt.com 633

Volume 4, Issue 10, October – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. RESULTS

We used some steps to get output. This image

segmentation are used to get a unique picture.

Fig 2:- Voxel

D. 3D Modelling
In this project, the real 2D medical images were used
as the input data. Required number of DICOM images of
human parts will be in-putted to the system to get a high
accuracy 3D model of the given DICOM images set.

IV. MARCHING CUBES ALGORITHM

Fig 5:- Processing of Volume
The Marching Cubes (MC) algorithm by Lorensen
and Cline is most used algorithm for extraction of iso
Canny edge identification is connected on the first
surface out of volumetric data. This algorithm produces a
picture so we clearly identify border edge. While, at that
triangle mesh by computing iso-surfaces from discrete data.
point Gaussian calculation is used on the primary picture to
a. By drawing all these triangles, wepcan build a three get the perception y. The outcomes discovered uncovers the
dimensional representation of the CT-Scans [17] [18].
underlying picture which are come about because of the k-
means have morphological openings, and not safeguard the
watchful edges, for example these picture are not smooth
enough. In this way, the HMRF refines picture out all of
these disservices .The last outcome is segment effect area
and calculate volume.

Fig 3:- 3D rendering by marching cube algorithm

Fig 6:- Trained Graph of Voice Recognition

It detects system words. When the physician presses

the AI, in voice recognized function, the voice of the
physician is detected and converted in to text. The
specialized thing here button and his voice is converted to
audio through the system. Then it’s converted in to text. The
below table will lead us to measure the accuracy and the
efficiency of the voice recognition system. Table show the
test value and train data.
Fig 4:- Process of DICOM

IJISRT19OCT2218 www.ijisrt.com 634

Volume 4, Issue 10, October – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[7]. B. W. Y. Santosh K.Gaikwad, " “A Review on
Speech Recognition Technique”,," in International
Journal of Computer Applications (0975 – 8887),
November 2010., pp. Volume 10– No.3, .
[8]. kajal, "kajal," [Online]. Available:
https://www.kaggle.com/ucil/breast-caner-isonsin-
data..
[9]. A. A. J. P.-M. B. Duquette, "“3D segmentation of
abdominal aorta from CT-scan and MR images”,,"
Table 1:- Testing of Breast Cancer Computerized Medical Imaging and Graphics, vol.
36, pp. 294-303, 2012..
VI. CONCLUSION [10]. Y. B. M. a. S. S. Zhang, "“Segmentation of brain MR
images through a hidden Markov random field model
Radiologists can use lung cancer segmentation to and the expectation-maximization algorithm”,,"
make the 3D lung model and calculate the region of lung Imaging, IEEE Transactions on, vol. 20, pp. 45-57,
cancer without critically examining medical images. Voice 2001..
created in a scheme where the voice commands can [11]. Y. B. A. G. B. S. M. a. C. Zheng, "“Four-chamber
regulate the part of the system. It is unknown the-actual- heart modeling and automatic segmentation for 3-D
cause of-breast-cancer, but-early therapy can be a useful cardiac CT volumes using marginal space learning
way to prevent and detect breast cancer as well. Data and steerable features”," edical Imaging, IEEE
mining technology is-the simplest and easiest way to Transactions on, vol. 27, pp. 16681681, 2008..
predict whether it is recurrent or non-recurring breast [12]. M. H. V. a. B. R. Sonka, "“Image processing,
cancer. Data mining algorithms such as decision tree, analysis, and machine vision: Cengage Learning”,,"
support vector machine naïve Bayes give more precise 2014..
outcomes in many of the articles mentioned above. [13]. S. S. a. K. N. V. Al-Amri, "“Image segmentation by
using threshold techniques,," 2010. arXiv preprint
VII. FUTURE WORK arXiv:1005.4020.
[14]. D. T. A. S. C. a. H. B. Cooper, "Effect of voxel size
Only 3D lung pictures can be rendered by a on 3D micro-CT analysis of cortical bone porosity”,
developed scheme. It is possible to develop cancer in the Calcified tissue international,," vol. 80, pp. 211-219,
future by renderingp3D model of other body composition 2007.
and anatomies such as core. In addition, voice could be [15]. lg,"logical,"[Online].Available: ieeexplore.ieee.org
created into a scheme where voice commands can regulate [16]. K-nn, "K-NN," [Online]. Available:
the entire PAC scheme and Voice controlling is more https://www.researchgate.net/publication/242579096_
accurate and more function controlling, reducing manual A
operating time. Developing other related disease prediction. n_Introduction_to_Logistic_Regression_Analysis_an
d_ Reporting.
REFERENCES [17]. Sio, C. C., Ngan, M., Yi, J., & Chen, X. (2011).
Volume Rendering with Marching Cube
[1]. Health, "Healthdata," 2019. [Online]. Available: Algorithm.University of Southern California, date
http://www.healthdata.org/sri-lanka. Feb, 11.
[2]. Q. J.R, "“Induction of Decision Trees," Volume 1,, [18]. [Online].Available:indexive.com
March, 1986. [19]. J. Guo et al., “Revealing determinant factors for early
[3]. N. P. P. S. S. a. S. A. Ayushi Trivedi, "Speech to text breast cancer recurrence by decision tree,” Inf. Syst.
and text to speech recognition systems-Areview," Front., 2017.
DepartmentofComputerScience, NMIMS [20]. R.M..Sherekar and AnandPawar, "A MATLAB
University,Mumbai,India. Image processing approachfor reconstruction of
[4]. V. R. J. Daemen, "The block cipher rijindael", DICOM Images for Manufacturing of
Proceedings of the Third International Conference on customizedanatomical implants by using rapid
smart," 2000,. prototyping,"American journal of Mechanical
[5]. [Online]. Available: Engineering and Automation, vol. 1, no. 5, pp. 48-
https://www.aapm.org/meetings/amos2/pdf/35-9796- 53,2014.
86628-984.pdf. [6] B.-H. J. B. Lawrence Rabiner, [21]. Fuzzylogic
"Fundamentals of Speech Recognition". [Online]Available:https://www.scribd.com/book/1457
[6]. E. M. B. Raghavendhar Reddy, "”Speech to Text 18199/Fuzzy-Logic-with-Engineering-
Conversion using Android Platform”,," in Applications[Accessed:16-September-2016].
International Journal of Engineering Research and [22]. M Karuna and Ankita Joshi, “Automatic detection
Applications (IJERA) ISSN: 2248-9622, January - and severity analaysis of brain tumors using gui in
February 2013, pp. ,Vol. 3, matlab,” International Journal of Research in
Engineering and Technology, vol. 02, no. 10, 2013.

IJISRT19OCT2218 www.ijisrt.com 635

Volume 4, Issue 10, October – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[23]. Dr. Denham Pole “Electronic Patient Records in Sri
Lankan Hospitals”, Sri Lanka Journal of Bio-Medical
Informatics 2010;1(1):43-45 [Accessed: January 04
2015]
[24]. E.G Devine, S. A. Gaehde, and A. C. Curtis,
“Comparative Evaluation of Three Continuous
Speech Recognition Software Packages in the
Generation of Medical Reports,” J. Am. Med.
Informatics Assoc., vol. 7, no. 5, pp. 462–468, 2000.

IJISRT19OCT2218 www.ijisrt.com 636