Detection of Blood Vessels In The Human Body Using The Semantic Segmentation Model

Abstract

Our model focuses on the detection of blood vessels in retinal images, and it incorporates deep learning alongside image processing algorithms to produce accurate and efficient design [1]. Deep learning techniques are frequently utilized for medical image analysis because of the quick progress in both technology and medical knowledge.[2]. A modified UNET semantic segmentation neural network will be used in our study which contains 26 layers instead of 23 layer in original model this allows for better recognition of minor vessels by increasing the activation layer and optimizing the kernel size. A collection of retinal images will be used to train the model, The dataset used in the model contains 20 images of train and test with 20 mask for both of them and the size of image is 512x512, it is named DRIVE dataset and is used in multiple journals. This is one of two available dataset of fundus retinal images. The training is done with using the Adam optimizer and in batch size of 2. testing will be done to assess the neural network's performance. We will evaluate our methodology's correctness and efficiency based on the initial testing results. Throughout this research, it can be concluded that deep learning segmentation approaches can provide high efficiency as well as high accuracy in blood vessel detection compared with conventional pixel-based detection methods. Which we got 96.53% and it resulted in accurate mapping of both large and smaller blood vessels, we have achieved .8052 as our F1 score is which one of the top from previous studies. We have achieved highest MOIU from previous leader E-NET model i.e. .7744 for them and .8184 for us which is highest. The neural network can be used to diagnose other associated conditions including hypertension, diabetic retinopathy, and others since it can recognize the patterns and form of blood vessels. Furthermore, the neural network's great efficiency makes it possible for the algorithm to identify blood vessels in real time. Our work expands the applicability of deep learning techniques in the field of medical image analysis and holds the potential to transform the present approaches to blood vessel-related disease diagnosis and detection which can help in early detection of diseases such as glaucoma and multiple other cancerous and diabetic diseases.