Unknown power power point unknown power point

LIVER LESION DETECTION USING CT SCANS
From:
Nikitha C H [1AR19CS030]
Monika S [1AR19CS026]
Jayashree A Naik [1AR19CS017]
Vishwas Gowda H V [1AR19CS061]
Project Guide:
Prof. Srinivasa R

AGENDA
 Abstract
 Introduction
 Literature Survey
 Problem Identification
 Objectives
 Requirements
 Applications
 System Design
 Methodology
 Results
 Conclusion

ABSTRACT
 Liver segmentation is an essential prerequisite for liver cancer
diagnosis and surgical planning
 Traditionally, liver CT Scans are gone through manually by
radiologist in a slice-by-slice fashion. However, this process is
time-consuming and prone to errors depending on
radiologist’s experience.
 In this Project a CNN Based pyramid pooling Architecture is
used to overcome this problem by scanning CT Scans and
Filtering

INTRODUCTION
 According to the survey done in the year 2017, 800,000 new cases
for liver cancer arises every year making it the sixth most common
cause of death due to any cancer in middle- and lower-income
group.
 Liver cancer is the subsequent cause of death worldwide after lung
and breast cancer.
 CT scans with high resolution serve as an incentive test for liver
tumour detection.
 To Reduce the mortality rate clinical experts call for computer aided
detection of liver tumour.

LITERATURE SURVEY
YEAR TITLE/AUTHOR LIMITATIONS
2004
Liver cancer imaging: role of CT,
MRI, US and PET by Maria
Raquel Oliva and Sanjay Saini
It gives less accuracy.
2013
A Hybrid Forest Classifier along
with SVM by Frimmel , Basha.
Support Vector
Machine (SVM) based
Liver tumour
classifications are
achieved previously
with less precision
rate.

EXISTING SYSTEM
 Abdominal CT scans have been widely studied and researched
by medical professionals in recent years.
 Computer-aided automatic segmentation of the liver can serve
as an elementary step for radiologists to trace anomalies in the
liver.
Disadvantage :- Approach used is not appropriate . The liver
cancer disease is detected and confirmed through medical tests
through medical professionals.

PROPOSED SYSTEM
 We have explored deep learning techniques first and foremost
for the extraction of liver from the abdominal CT scan.
 The objective is to detect the liver lesion by using
Convolutional Neural Network(CNN). This is done by giving
CT Scans as input and then the predictions are made as
normal or abnormal liver and it also detects the percentage of
lesion in liver also gives a print out to user.

PROBLEM IDENTIFICATION
 As Population Growing across Country Many Diseases are
taking control over humans, and Inner organs like liver and
other are most effected one.
 It is difficult even for clinical experts to distinguish between a
healthy liver and tumoral liver due to undetectable differences
between them.

OBJECTIVES
 The objective is to detect the liver lesion by using
Convolutional Neural Network(CNN).
 This is done by giving image dataset as input and then the
predictions are made as normal or abnormal liver with the
tumour percentage with respect to Liver Size.
 We have explored deep learning techniques for the extraction
of liver from the abdominal CT scan.

REQUIREMENTS
 Hardware Requirements:
Processor type : Intel core i5 and above
Processor speed : Minimum 2.00 GHz and above
RAM : 6-10 GB
Hard Disk : 400 GB or more
Monitor : 800x600 or higher resolution
Keyboard : 110 keys enhanced
 Software Requirements:
Operating System : Windows XP,7,8,10
Technology : Python
Front End : Tkinter
IDLE : Python 2.7 or 3.0 or higher, Tensor flow
Database : MySQL
Software used: Tensor Flow

APPLICATIONS
 Time Saving and effective Results.
 It can be easily Understood By Lab Technicians.
 As output shows clear affected part of liver, it is visually
effective to watch the tumour part.
 Reduces a time of Consulting Doctor for analysing CT Scans.

USER INTERFACE
 In this User interface is a Web page Hosted on server.
 In this User has to Provide their CT Scans as a input to predict
the CT Scans are normal or abnormal.
 After performing a Implementation output will be displayed on
the web page itself by labelling weather it is abnormal or not.
 It also shows the lesion percentage in the liver and gives
patient issue copy.

CNN
 Convolutional neural network is class of artificial neural network.
 Applied to analyze images.
 Involves processing of pixel data.
 Its built in convolution layer reduces high dimensionality of images
without losing information.

CNN LAYERS
 Input Layer
 Convolution Layer
 Activation Function Layer
 Pooling Layer
 Fully Connected Layer

 Input Layer
 This layer holds the raw input images.
 It is composed of artificial input neurones and brings the
initial data into the system for further processing
 Convolution Layer
• Convolution preserves the relationship between pixels by
learning image features using small squares of input data.
• Where the given image is converted to 2D or 3D based on
the requirements. Here 3D image is converted to 2D

 Active Function Layer
• This layer will apply element wise activation function to the
output of convolutional layer.
• Height,width of the given image is compared and computed
with activation functions.
 Pooling Layer
 Pooling Layer is building block of a CNN. Its function is to
progressively reduce the spatial size of the representation to
reduce the amount of parameters and computation in the
network.
 Here the time complexity is reduced along with size. Where
it reduces the dimensionality of given image with presence
of important information.

 Fully Connected Layer
 This layer takes the output of pooling and predicts the label
to describe the image.
 In this layer we fastened our matrix into vector and feed it
into a fully connected layer like neural network.
 With fully connected layer we combined these features
together to create a model.
 Finally the respective required model is created using
above layers.

SYSTEM DESIGN
 SYSTEM ARCHITECTURE

IMPLEMENTATION
METHODOLOGY
 Dataset Implementation
 The labeled training sets of the LiTS17 and SLiver07
datasets are used. The LiTS17-Training dataset consists
of 131 sets of 3D abdominal CT scans
 In our project the data is collected by website named
kaggle.
 Also 1500+ images of training data set and 500+ images
for Testing Dataset.

 Image-Processing
 We are resizing the Dataset image to 512x512 Dimensions
Before Passing it to the Model.
 This is because Segmentation of images having 512x512
dimensions is more Faster on Modern GPU i.e Less than a
second.
 It removes irrelevant organs from CT scans, the
Converts intensity is transformed to [-200, 200] to
enhance the contrast and clarity of the image.

 Dataset Augmentation
 variety of rigid and elastic transformations were utilized, including: scaling
the image between and 1.2 with a 50% probability.
 rotating the image between 0 degrees and 30 degrees with a 30%
probability
 Using Algorithm (Pyramid Based U-Net)
 In this Project we have used Pyramid Based pooling U-net Architecture
which is a CNN Architecture
 In this Architecture uploaded input is Filtered Using CV2 Convert and
Filtered Imaged is Used for Further Evaluation

 Evaluation
 Many Evaluation Metrics are Used while Classifing Input that are
Dice, volume overlap error (VOE), relative volume error (RVD),
average symmetrical surface distance (ASSD), and Maximum
Surface Distance (MSD).
 For Dice and VOE, the larger the value, the better the
segmentation result, while ASSD, RVD and MSD are the opposite
 Prediction
 In this project we get the part of liver image that has been
predicted by Model as output.
 We will get to know that what percentage of liver is affected with
Malignant Cells (Tumour).

INPUT DATA SAMPLE
Sample of a normal image
dataset
Sample of abnormal image
dataset

RESULTS AND SCREENSHOTS
Fig:- Signup Page

Fig:- Interface to upload CT Scans

Fig:- Result Page showing Filtered Image and
Predicted Images with infected percentage

Fig:- Patient Issue copy with all Results related to CT
Scans for further Treatment

CONCLUSION
 In this work, we have proposed a pipeline for segmentation of
liver and its lesion using different architectures of
convolutional networks, other vital information about the
tumour could also be extracted like the precise dimensions of
the tumour detected.
 we believe that our proposed method could be used to serve
as preliminary analysis for radiologists

Unknown power power point unknown power point

More Related Content

Similar to Unknown power power point unknown power point (20)

Recently uploaded (20)

Unknown power power point unknown power point