Figure 3
From: Semi-supervised segmentation of retinoblastoma tumors in fundus images
Results of segmentation refinement, based on the second approach, where the model is trained using original images as input data and manually edited outputs of unsupervised method as output data, while no thresholding step is performed. This approach is equally fast compared to its predecessor, but it can provide more detailed information about the height of detected tumors. The method can be further improved by increasing the resolution of input images and enhancing computational speed, resulting in the most accurate and detailed segmentation of retinoblastoma in this study. Nevertheless, this method may face challenges in detecting the optic disc as an anomaly compared to the introduced supervised approach.
