Semantic Segmentation of Retinal Blood Vessel via Multi-scale Convolutional Neural Network

Autor:	SM Mazharul Islam
Rok vydání:	2020
Předmět:	Computer science business.industry Tsallis entropy Glaucoma Pattern recognition Diabetic retinopathy Fundus (eye) medicine.disease Convolutional neural network medicine Adaptive histogram equalization Segmentation Artificial intelligence Sensitivity (control systems) business
Zdroj:	Proceedings of International Joint Conference on Computational Intelligence ISBN: 9789811536069 IJCCI
Popis:	Segmentation of retinal blood vessel is one of the key stages to automate diagnosis and detection of diseases such as diabetic retinopathy (DR), glaucoma, hypertension, age-related macular degeneration (AMD), etc. In this paper, semantic segmentation method is presented to extract the blood vessels from retinal fundus images using a novel multi-scale convolutional neural network. Features are extracted from raw fundus images via Tsallis entropy, adaptive histogram equalization and standard deviation. Then, a multi-scale convolutional neural network is applied on these features to semantically segment the blood vessels. The algorithm is trained only with the training set images of DRIVE database and then tested on both test-set images of DRIVE database and healthy patient images of HRF database to evaluate its performance. During the whole evaluation process, the network and its weights have remained fixed. The results show that the proposed method achieves an accuracy, sensitivity, specificity of 95.38%, 79.23%, 97.73%, respectively, on DRIVE database and 94.03%, 77.52%, 96.07%, respectively, on HRF database.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::1f6fa1116516477a5e871dc0c709be3d https://doi.org/10.1007/978-981-15-3607-6_18 Zobrazit plný text záznamu