Manuscript received January 23, 2026; revised February 24, 2026; accepted April 15, 2026; published September 11, 2026.

Abstract—Retina vascular segmentation exerts major part in the identification of age-related macular disease and diabetic retinopathy. Accurate retinal vessel segmentation depends on a range of enhancement techniques and deep learning models. Fundus images are primarily used for diagnosing eye diseases. High-quality visual information supports highly accurate clinical judgment. Conventional deep learning models commonly impair accuracy due to intricacy. However, the quality of the image frequently affects various factors such as noise. To overcome these limitations, we introduced a Stacked Efficient Depth (SED) encoder-decoder framework for retinal vascular segmentation. In this framework, we used four blocks, such as Gaussian Convolutional Residual (GCR), SED encoder, Transformer, and SED decoder. The GCR block introduces stochastic noise into the input during training, and the model develops increased robustness against various input disturbances. The SED encoder applies spatial filtering independently to each input channel. Moreover, a Vision Transformer block is used in the bottleneck to fuse the interest region detection of SED with global reasoning. The SED decoder block uses depthwise convolution to minimize the number of parameters. The proposed model’s effectiveness is validated through both objective and subjective evaluations on publicly available datasets. Our proposed method demonstrates superior performance in evaluation metrics to existing approaches.

Keywords—transformer, retinal vessel enhancement, stacked efficient depth, Gaussian convolutional residual
 

Cite: Srinivas Rangu and Nagaraj Yamanakkanavar, "Translight: Transformer Lightweight Model for Retinal Vascular Segmentation Using Fundus Images," Journal of Image and Graphics, Vol. 14, No. 5, pp. 772-786, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).