Liver Tumor diagnosis Desktop Application using Deep Learning and Challenging Dataset

Project Title

Liver Tumor diagnosis Desktop Application using Deep Learning and Challenging Dataset

Project Area of Specialization Artificial IntelligenceProject Summary

With the passage of time diseases are becoming more complicated, advanced and difficult to
diagnose at an early stage such as liver disease. Liver disease is difficult to diagnose at early
stages because of subtle symptoms. Since nobody can function or live a healthy life without a
liver. Our aim of this project is to improve diagnosis of liver at early stages, save time of doctors
and to make early treatment possible. Our web based application works on the CT scan and MRI
images of liver and detects any abnormalities such as clot , expansion of tissues , tumour etc by
using Machine learning methods , pre defined Data sets , advanced techniques of image
processing and AI . This web App processes the 2D images, giving detailed list of symptoms or
abnormalities and diagnosis the liver tumour. Standard datasets are scanned via this web app, and
uploaded for the patients so that they can get those scanned datasets through this web app. It
helps patients to start early possible treatments and cure the disease. As privacy is concerned, the
reports remain confidential between patients and appointed doctor throughout this web App.

Project Objectives

The objectives are as follows:

1. To review state of-the-art liver tumor diagnosis methods for early liver tumor diagnosis
2.  To develop an intelligent system using deep learning method for liver tumor diagnosis
3.  To improve the performance of proposed system of state-of-the-art liver tumor dataset
4.  To facilitate doctors & patients with early diagnosis by providing user-friendly application

Project Implementation Method

Experimental Setup:
At first, we will do the preprocessing:

1.  Windowing the HU values to [-100, 400]
2.  Reduce width and height dimension to (168, 168) for fast iteration
3.  Set scan spacing to 2mm for fast iteration
4.  Max-min normalization

Training
 Used a 3d UNet model as defined by the script for training. The model developed
with reference on the 2d UNet model with 3d counterparts.
 Predict liver segmentation and lesion segmentation at the same time
 Used patch size of (168, 168, 16)
Tools/Technology:

1. Jupyter Notebook / Google collab ( Execution of python code )
2.  Python ( Programming Language )
3.  Qt designer ( For developing User Interface )

Our proposed segmentation workflow is depicted in diagram. The workflow consists of three major
steps
1. Data Preparation:
In Figure (1) The first step deals with data preprocessing and preparation for the neural network
segmentation. First, the Hounsfield unit values were windowed in the range [?100, 400] to exclude
irrelevant organs and objects. After that We will increase contrast through histogram equalization.
And then the contrast within the liver has been enhanced to allow better differentiation of abnormal
liver tissue.
2. Cascaded Fully Convolutional Neural Networks
In a second step two cascaded fully convolutional neural networks first segment the liver and then
lesions within the liver region-of-interest (ROI).

3. In the final third step, the calculated probabilities of CFCN will be refined using a dense 3D
conditional random field to produce the final segmentation result.

Overview of the proposed image segmentation workflow for training and testing. As the first step the
CT/MRI volumes are preprocessed with either HU-windowing or N4 bias correction. During the
training phase the training data is augmented to foster the learning of invariance against noise and
deformations in medical data. The CT/MRI volumes are trained after pre-processing and data
augmentation in a cascaded fully convolutional neural network (CFCN). A first FCN segments the liver
from abdomen CT/MRI scans. This segmented liver region of interest ROI is the input for a second FCN, that segments lesions from the given segmented liver ROI. To gain the final segmented volume
is refined afterwards using a 3D conditional random field 3D CRF

Benefits of the Project

Benefits:

1. It facilitates the doctors and as well as patients in early diagosis of tumor if any so, that it can be treated easily.

2. The desktop Application saves time of doctors and patients and patients don't have to take appointments for a checkup.

3. As, diagnosing and treating tumor  by doctors manuallyis a costly method so, this system can reduced that cost.

Technical Details of Final Deliverable

The proposed system is a solution for all users of the system to diagnose Liver Tumor.
To diagnose the disease in early stages is a new growing need all around the world speciallyto
achieve higher sensitivity and specificity as compared to traditional methods. The system will
analyze CT scans to diagnose Liver Tumor. The system will allow all users of the system to input
a CT scan on the system and the system will analyze the CT scans and will provide the results.
The system will reduce the cost and time to diagnose these diseases. The results produced will
be more accurate than the traditional radiologists. The system will serve humanity and save lives.

Tools/Technology:
1. Jupyter Notebook / Google collab ( Execution of python code )
2.  Python ( Programming Language )
3.  Qt designer ( For developing User Interface )

Final Deliverable of the Project Software SystemCore Industry MedicalOther Industries Health Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	65000
Graphic Card (Gtx 1060 6gb)	Equipment	1	65000	65000