Deep Learning Based Framework for Colorectal Polyp Classification

Project Title

Deep Learning Based Framework for Colorectal Polyp Classification

Project Area of Specialization Software EngineeringProject Summary

The proposed project will be a web-based artificially intelligent application that will help doctors to detect colorectal cancer by taking the images of detected polyps produced by endoscopy and tell the end-user which kind or class of polyp is present in the image through deep learning. There will be three types of end-users interacting with the system including:

• Patients, who will only be able to view the results of his/her endoscopy tests given by the doctor.
• Doctor/Physician, who can give the image input to the system and analyze the result. Furthermore, will be able to see the record of any patient.
• Admin, who can manage all the current situations occurring in the whole system. Admin will register Doctors and Patients. Moreover, Admin will be able to view the report of all registered and patients. Admin will set appointment dates and times of patients with the doctors.

Our Project will take colonoscopy images of polyp in the large intestine, it will process these images with implemented deep learning algorithms and will generate a report accordingly. It will analyze the images and will correctly identify the first three types of polyp as adenomatous, serrated, and hyperplastic polyps. The existing work in classifying two types of polyp and is not commercially available to every doctor and physician that can get a classification report from an AI trained system. Moreover, the researcher trained their model with many algorithms on different images taken in endoscopy. Existing work shows that the Convolutional Neural Network (CNN) algorithms perform better in extracting features from images. So, we will train our model with CNN on a large data set to get more accurate results. Based on results generated by the system, doctors will be able to detect cancer.

Project Objectives

• To enable doctors to detect colorectal cancer in their patients based on artificially intelligent mechanism of analysis of endoscopy reports easily.
• To develop a system that can classify polyp into three types as adenomatous, serrated, and hyperplastic.
• To develop a system that will assist doctors and physicians in the classification of the polyp by providing a facility to crosscheck their remarks with system generated report of classification.
• To provide a system that provides facility for the patient to check their daily report and discuss the internal condition of the body directly with the doctor.
• According to a report of NICE classification of polyps, the accuracy of 10 physicians on 735 polyps was 76.7 percent. Our system will use the Deep learning algorithm and provide more accuracy and facilitate the doctors.

To provide a system that helps in reducing the death rate of patients by informing them in the early stages of colorectal cancer.

Project Implementation Method

In Software Development Life Cycle, many software process models could be followed to achieve the best results. These models can be implemented depending upon the nature of the software. Each model has its advantages and disadvantages. Our system is critical and it involves risks that result must be accurate, otherwise, we can lose somebody’s life. So, we will follow the Spiral model.

Methodology:

The spiral model is known as the Meta model because it subsumes all the other SDLC models. This model can handle risk. It is flexible to handle as requirements change later. If we want to change or add a new feature, it allows us to do it at any phase of development. Each module of the system will pass through the spiral development life-cycle including Plan, Design, Develop, Test, Release, Feedback, and necessary changes will be made.

For the classification of polyps through colonoscopy images, we will use a deep learning algorithm to extract features and perform analysis from images. There are many deep learning algorithms available like Convolutional Neural Network (CNN), Stack Sparse Autoencoder (SSAE), etc. As found in the literature review that Convolutional Neural Network (CNN) and Deep Neural Network (DNN) gave more accuracy in processing images. We will test these algorithms according to our data set and we will implement the algorithm producing better accuracy.

Phase 1: In the first phase, we gather data set for training of the model and identify the features and milestones. We will analyze the architecture and draw UML diagrams for a better understanding of the system.

Phase 2: The second phase will start with designing web applications in PHP. Which included the front end of the Admin module, Patient module, and Doctor module. In this phase, a database for storing records will also be implemented.

Phase 3: This phase is large and time-consuming. In this phase, we will implement above mention deep learning algorithms and test the accuracy of each algorithm on our data set.

Model Diagram

Benefits of the Project

This project will be specially designed for physicians and doctors who classify the family of polyp found in patient’s colon by providing them the facility to crosscheck their remarks with the artificially trained software system having a much higher accuracy rate.

 It will be providing a great facility to the patient that they can instantly view their reports and he/she will be able to track his/her history and can keep a record of his results.

Doctors will view the patient’s reports and previous medical history so that the Doctor will get a better understanding of the patient’s disease and will start the treatment at pre-stages of Colorectal cancer. Only the treatment can stop the conversion of polyp into cancerous malignant and can save a patient’s life.

Technical Details of Final Deliverable

The final deliverable of the project will be a web application along with a documentation manual having all the technical details of the application.

Tools:

Xamp: Localhost and database.

Visual studio code: Development of web application.

Anaconda: Model training using a deep learning algorithm.

Languages:

HTML, CSS, JAVASCRIPT, and PHP: Web application development.

Python: Model implementation and training.

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 People, Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	67500
2 TB Hard Drive	Equipment	1	25000	25000
OTP Gateway	Equipment	1	15000	15000
Domain	Equipment	1	10000	10000
Cloud Hosting	Equipment	1	10000	10000
Printing (Documentation)	Miscellaneous	1	7500	7500
Printing	Miscellaneous	0	0	0