Polyps Detection Using Image Segmentation Algorithms

Project Title

Polyps Detection Using Image Segmentation Algorithms

Project Area of Specialization Artificial IntelligenceProject Summary

Identification of polyps in endoscopic images is critical for the diagnosis of Colorectal Cancer (CRC). Finding the exact shape and size of polyps requires the segmentation of endoscopic images. Accurately detecting and segmenting polyps can help endoscopists to remove the abnormal tissue through medical surgeries and procedures, thereby decreasing the chances of polyps developing into cancer. Since high human dependency is observed in Colonoscopy (COL) examination, this increases the chance that polyps can be missed during this procedure thus increasing the possibility of cancer development. The proposed solution is to develop a Computer-Aided Detection (CAD) system that will counter this problem and provide another point of view to the doctors, so that their negligence may not result in the death of an individual.

Project Objectives

Colonoscopy is heavily dependent on the experience of endoscopists this results in some polyps being missed during the exam. To lower the percentage of missed polyps and to make Colonoscopy partially independent from an endoscopist, we will develop a Computer-Aided Detection (CAD) System which performs image segmentation on a medical dataset which will ensure that a maximum number of polyps are detected accurately. Our project will be developed to help the clinicians have a second opinion so that if the doctor missed any polyps throughout the procedure, this CAD system should be able to help the doctor rectify their mistake and possibly save a patient’s life.

Project Implementation Method

We will implement a modified version of Resunet++ as our deep learning approach. The ResUNet++ architecture was employed which uses the encoder and decoder structure for semantic segmentation. Pyramid pooling was used as a bridge between the encoder and decoder block. The encoder block contains residual units that take advantage of skip connection in a neural network. The skip connection allows training a deep neural network without degrading the performance. Squeeze and excitation blocks were used which ensure that the channel output features are weighted equally. The attention mechanism is used in the decoder block. The attention mechanism is useful in making a pixel-wise prediction. In semantic segmentation, an attention mechanism is used to give attention to each pixel of an image which can then be used to predict the pixel level.

This Atrous Spatial Pyramid Pooling (ASPP) block in ResUNet++ was implemented using depth-wise separable convolution as well as replaced with Deep Atrous Spatial Pyramid Pooling (DASPP) module from in a separate experiment. The implementation of depth-wise separable convolution is done by applying kernel on input at channel level. The output from here is passed through the pointwise convolution with a 1x1 kernel. The application of depthwise convolution results in fewer GFLOPs and parameters.DASPP was implemented to see if going deep in the network improves performance on polyps segmentation.

Benefits of the Project

Missed polyps may develop into CRC so they should be identified and removed as early as possible, so a  CAD System would be beneficial for the patients and the doctors in ensuring that they never miss polyps in the first place. 

Our project will be developed to help the clinicians have a second opinion with our proposed model  which will do pixel-wise image segmentation to detect polyps in the gastrointestinal (GI) tract using modern DL based Convolutional Neural Networks (CNN) so that if the doctor missed any polyps throughout the procedure, this CAD system should be able to help the doctor rectify their mistake and possibly save a patient’s life

Technical Details of Final Deliverable

The final deliverable includes a website and a mobile app for the clinicians that will allow them to identify polyps from the images obtained through endoscopy.

The pipeline will be as the image is input from the frontend via mobile or web then the image is segmented through our proposed deep learning architecture and the resultant segmented image is then shown to the clinicians. In this way, the minor polyps that can't be seen through the naked eye will be segmented using our deep learning model.

Final Deliverable of the Project Software SystemCore Industry HealthOther Industries IT , Medical 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)	48000
GeForce GTX 1060 - 6GB GDDR5 - Twin X2 192Bits	Equipment	1	48000	48000