Blood Vessel Segmentation in Retinal Images using PCA and AI

Abstract

Medical Image Processing plays a vital role in the early diagnosis and treatment of ophthalmic diseases. Retinal blood vessel analysis is one of the most important tasks in retinal image processing because the structure and condition of retinal vessels provide valuable information regarding diseases such as Diabetic Retinopathy, Glaucoma, hypertension, and macular degeneration. Manual examination of retinal images by ophthalmologists is time-consuming, labour-intensive, and subject to human error. Therefore, there is a strong need for an automated and intelligent system that can accurately segment retinal blood vessels and assist in disease prediction This project proposes an advanced framework titled Blood Vessel Segmentation in Retinal Images using PCA and U-Net with AI-Based Disease Prediction. The proposed system integrates Principal Component Analysis,

Convolutional Neural Network, and U-Net to achieve efficient and accurate retinal vessel segmentation. The system is designed to improve the quality of retinal image analysis while reducing computational complexity and enhancing diagnostic capability Initially, retinal fundus images are collected from standard datasets and pre-processed using grayscale conversion, resizing, and normalization techniques.

These preprocessing steps improve image quality and prepare the images for segmentation. A patch-based PCA technique is then applied to extract important vascular features and reduce unnecessary information from the retinal images. PCA enhances the visibility of blood vessels by preserving significant features while suppressing noise and redundant data. The transformed PCA feature maps are combined with original retinal images to form enhanced multi-channel inputs for deep learning models The proposed framework utilizes two segmentation approaches. The first approach uses a CNN model for basic vessel detection and segmentation. CNN automatically learns hierarchical image features such as edges, textures, and vessel structures. However, standard CNN models may fail to capture thin vessels and fine details. To overcome this limitation, the second approach integrates PCA-enhanced inputs with the U-Net architecture. U-Net consists of encoder and decoder layers connected through skip connections, enabling the model to capture both global contextual information and fine-grained vessel details. This hybrid PCA + U-Net model produces more accurate and refined vessel segmentation results compared to traditional methods After segmentation; post-processing operations such as thresholding, median filtering, and morphological processing are applied to improve the quality of the segmented masks. These operations remove noise, smooth vessel boundaries, and enhance vessel continuity. The segmented retinal vessels are then analyzed to calculate vessel density, which is an important indicator of retinal health. Based on vessel density values and regional irregularity analysis, the system predicts the severity of retinal diseases and classifies conditions into healthy, mild, moderate, or severe categories An additional feature of the proposed system is the integration of an AI-based advisory module that generates explanations, possible causes, and precautionary suggestions for the detected retinal condition. This intelligent module improves system interpretability and helps users better understand diagnostic outcomes. The entire application is implemented through a user-friendly graphical user interface developed using Python Tkinter, allowing users to upload datasets, train models, perform predictions, and visualize outputs interactively