Browsing by Author "Rafid Masrur Khan"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Open Access
    Digital Retinal Images for Vessel Extraction Using Deep Learning Approach
    (North South University, 2021) Md. Fahad Mojumder; Adnan Ahmed Saif; Rafid Masrur Khan; Mahdy Rahman Chowdhury; 1712145642; 1620611042; 1520862642
    Retinal vessel extraction using retina image is a deep learning based semantic segmentation process by which we will train machine using retina images and extract the vessels from it. U-net model is best for this semantic segmentation and we use this model which help machine to learn from the images in order to provide better accuracy. Retinal vessel extraction is a process by which a physician can detect the anomaly in retina and this can prevent blindness if found early stage. In this research we have tried to implement U-net model on different types of datasets and tried to achieve better accuracy. We have used different techniques like data preprocessing, data augmentation, image slicer, and contour to achieve better result. Data preprocessing is used for proper feature selection and feature extraction. We have removed extra noise from the input images in order to achieve better result. Selecting important portion and removing irrelevant portion of input features is the main aim of data preprocessing. Data augmentation is needed to increase the number of input features. We have used data augmentation by modifying input pictures and increase the number of images. This increase number of slightly modified data helps the model to learn better in learning stage. Then image slicing is used to slice one image in multiple segments. These multiple segments help to highlight important features of the image. Thus, the architecture gets better and more concentrated features for learning stage. At last, the contour is used to remove irrelevant features like the black background of our input pictures and make the boundary continuous with the outline of the relevant features. Thus, we have tried to find out how our model works on publicly available three datasets DRIVE, STARE and CHASE_DB1 and achieve better result in our project. We tried to do semantic segmentation using these datasets and tried to overcome the obstacles we found during this process.