Weakly-Supervised Vessel Detection in Ultra-Widefield Fundus Photography Via Iterative Multi-Modal Registration and Learning

Abstract

<div>We propose a deep-learning based annotation efficient framework for vessel detection in ultra-widefield (UWF) fundus photography (FP) that does not require de novo labeled UWF FP vessel maps. Our approach utilizes concurrently captured UWF fluorescein angiography (FA) images, for which effective deep learning approaches have recently become available, and iterates between a multi-modal registration step and a weakly-supervised learning step. In the registration step, the UWF FA vessel maps detected with a pre-trained deep neural network (DNN) are registered with the UWF FP via parametric chamfer alignment. The warped vessel maps can be used as the tentative training data but inevitably contain incorrect (noisy) labels due to the differences between FA and FP modalities and the errors in the registration. In the learning step, a robust learning method is proposed to train DNNs with noisy labels. The detected FP vessel maps are used for the registration in the following iteration. The registration and the vessel detection benefit from each other and are progressively improved. Once trained, the UWF FP vessel detection DNN from the proposed approach allows FP vessel detection without requiring concurrently captured UWF FA images. We validate the proposed framework on a new UWF FP dataset, PRIMEFP20, and on existing narrow field FP datasets. Experimental evaluation, using both pixel wise metrics and the CAL metrics designed to provide better agreement with human assessment, shows that the proposed approach provides accurate vessel detection, without requiring manually labeled UWF FP training data.</div>

RAVIR: A Dataset and Methodology for the Semantic Segmentation and Quantitative Analysis of Retinal Arteries and Veins in Infrared Reflectance Imaging

“…The ROSE dataset [16] provides 117 Optical Coherence Tomography Angiography (OCTA) images of 304 × 304 pixel resolution from 39 subjects suffering from Alzheimer's disease and, with regard to the Fluorescein Angiography (FA) modality, the RECOVERY-FA19 dataset [17] contains 8 ultra-wide-field images of size 3900 × 3072 pixels from patients participating in the Intravitreal Aflibercept for Retinal Non-Perfusion in Proliferative Diabetic Retinopathy trial [20]. Another FA dataset is the PRIME-FP20 dataset [18], comprising 15 images with a resolution of 4000 × 4000 pixels acquired from patients with diabetic retinopathy.…”

Learning-Based Algorithms for Vessel Tracking: A Review

“…The authors assumed that the probability map can better explain the uncertainty and subjectivity of detecting vessels, especially those at the edge locations appearing in the ground truth data. To formulate the vessel segmentation as a style-transfer problem, Ding et al (2020b) used the binary probability maps as the tentative training data and the style targets. Considering that shallower side-outputs capture rich detailed information, and deeper side-outputs have highlevel but fuzzy knowledge, Lin et al (2019) outputted the feature maps of each intermediate layer using VGGNet (Simonyan and Zisserman, 2015).…”

Student Becomes Decathlon Master in Retinal Vessel Segmentation via Dual-teacher Multi-target Domain Adaptation

“…These two modalities can both deliver precise representations of vascular structures within the retina, making them popular for the diagnoses of eye-related diseases. These years, advance in the imaging technologies brings new ophthalmic modalities including optical coherence tomography (OCT) [10], widefield fundus [4], and photoacoustic images [9]. These new modalities also provide important information of RVs and are likely to enhance the disease diagnosis accuracy.…”