CLASSIFICATION OF GLAUCOMA STAGES USING DEEP CNN WITH 2D- COMAPCT-VMD IN FUNDUS IMAGES
- Version
- Download 11
- File Size 543.99 KB
- File Count 1
- Create Date 26 April 2023
- Last Updated 26 April 2023
CLASSIFICATION OF GLAUCOMA STAGES USING DEEP CNN WITH 2D- COMAPCT-VMD IN FUNDUS IMAGES
N. Suneetha1, T. Siva Sankar2, K. Anuradha3, P. Aasif4, P. Sri Keerthana5, P. Uma Satish6 Asst.Professor, Sir C R Reddy College of Engineering1
UG Scholars, Sir C R Reddy College of Engineering 2,3,4,5,6
ABSTRACT: Diabetes has a consequence called glaucoma that can result in blindness. Using manual screening procedures to identify glaucoma in its early stages is costly and time-consuming. The high-level characteristics are acquired using a CNN model. The first layer of a pre-trained CNN model is reset using the Region of Interests (ROIs) of lesions retrieved from the annotated fundus pictures. After then, the model is adjusted such that the low-level layers pick up on the regional architecture of the lesion and healthy areas. In order to extract discriminatory characteristics from the fundus pictures unsupervised, we substitute the fully connected layer (FC), which encodes high-level features that are global in scope and domain-specific, with a new FC layer by considerably reducing the model complexity in this stage, the overfitting issue that determines whether diabetes is present is avoided. To create the final segmented image of the blood vessels in the Fundus image, the categorization for each pixel in the image is merged to determine whether they are normal or aberrant. In our work, the accuracy obtained for the proposed methodology is better.
INDEX TERMS:
Diabetes, Glaucoma, a CNN model, Region of Interests (ROIs), Fully-Connected (FC) Layer, Fundus Image.
Download