Fundus image quality enhancement for blood vessel detection via a neural network using CLAHE and Wiener filter

• Published in: Research on biomedical engineering Vol. 36; no. 2; pp. 107 - 119
• Main Authors: dos Santos, Jucelino Cardoso Marciano, Carrijo, Gilberto Arantes, de Fátima dos Santos Cardoso, Cristiane, Ferreira, Júlio César, Sousa, Pedro Moises, Patrocínio, Ana Cláudia
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.06.2020
• Subjects: Biomaterials; Biomedical Engineering and Bioengineering; Biomedical Engineering/Biotechnology; Engineering; Original Article; Fundus image; Blood vessel segmentation; Wiener filter; Neural network; CLAHE
• ISSN: 2446-4732; 2446-4740
• DOI: 10.1007/s42600-020-00046-y

Purpose Blood vessel segmentation is the most important step for detecting changes in retinal vascular structures in retinal images. While these images are widely used in clinical diagnosis, they are generally degraded by noise and are limited by low contrast. In this paper, we address the problem of improving fundus image quality for blood vessel detection. Methods We used contrast limited adaptive histogram equalization (CLAHE) to improve contrast and the Wiener filter for noise reduction. A multilayer artificial neural network was used to optimize the values from CLAHE and the Wiener filter for blood vessel segmentation. Furthermore, several training and classification rounds were performed (3240, with 200 epochs each), using a combination of CLAHE and Wiener parameters and a fixed network configuration. Results The proposed methodology was tested in the DRIVE database, achieving accuracy , sensitivity , and specificity values of 0.9505, 0.7564, and 0.9696, respectively. Conclusion The results were encouraging for almost all metrics and comparable to those of state-of-the-art blood vessel segmentation processes. Therefore, the parameter set effectively improved the fundus image quality for blood vessel segmentation, relative to the classification. These results are important since the more precise the segmentation step is, the greater the chances are of building a robust and specialized diagnostic system.