cs.CVMar 6, 2026

Vessel-Aware Deep Learning for OCTA-Based Detection of AMD

Authors: Margalit G. Mitzner, Moinak Bhattacharya, Zhilin Zou, Chao Chen, Prateek Prasanna

Organizations: Stony Brook University, Stony Brook, NY, USA

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Age-related macular degeneration (AMD) is characterized by early micro-vascular alterations that can be captured non-invasively using optical coherence tomography angiography (OCTA), yet most deep learning (DL) models rely on global features and fail to exploit clinically meaningful vascular biomarkers. We introduce an external multiplicative attention framework that incorporates vessel-specific tortuosity maps and vasculature dropout maps derived from arteries, veins, and capillaries. These biomarker maps are generated from vessel segmentations and smoothed across multiple spatial scales to highlight coherent patterns of vascular remodeling and capillary rarefaction. Tortuosity reflects abnormalities in vessel geometry linked to impaired auto-regulation, while dropout maps capture localized perfusion deficits that precede structural retinal damage. The maps are fused with the OCTA projection to guide a deep classifier toward physiologically relevant regions. Arterial tortuosity provided the most consistent discriminative value, while capillary dropout maps performed best among density-based variants, especially at larger smoothing scales. Our proposed method offers interpretable insights aligned with known AMD pathophysiology.

Figures & tables

Figure 1: Overview of the proposed pipeline vessel segmentations are used to generate tortuosity or density heatmaps (multi- σ\sigma ), which are then separately fused with the full OCTA projection via multiplicative weighting and classified using a ResNet-18 backbone for AMD detection.
Figure 2: Attention heatmaps. Segmentation labels, tortuosity dot maps showing relative tortuosity of each segment, high tortuosity (red hotspots indicate high tortuosity) and low vessel density heatmaps (red hotspots indicate low density).
MethodBal AccSpecSensAUC
No Attention0.86 ±\pm 0.070.87 ±\pm 0.120.86 ±\pm 0.120.90 ±\pm 0.08
Table 1: Baseline performance without attention. Values are mean ±\pm standard deviation.
  High Tortuosity External AttentionLow Density External Attention
Vesselσ\sigmaBal AccSpecSensAUCBal AccSpecSensAUC
Artery0.020.83 ±\pm 0.060.84 ±\pm 0.040.82 ±\pm 0.150.87 ±\pm 0.090.76 ±\pm 0.130.69 ±\pm 0.210.83 ±\pm 0.130.76 ±\pm 0.12
0.040.79 ±\pm 0.070.81 ±\pm 0.070.77 ±\pm 0.160.83 ±\pm 0.110.78 ±\pm 0.040.85 ±\pm 0.100.72 ±\pm 0.130.81 ±\pm 0.06
0.060.83 ±\pm 0.070.81 ±\pm 0.120.84 ±\pm 0.150.85 ±\pm 0.110.74 ±\pm 0.020.89 ±\pm 0.080.58 ±\pm 0.090.78 ±\pm 0.05
0.080.85 ±\pm 0.100.81 ±\pm 0.160.89 ±\pm 0.120.89 ±\pm 0.070.72 ±\pm 0.050.89 ±\pm 0.090.56 ±\pm 0.080.71 ±\pm 0.06
Vein0.020.83 ±\pm 0.080.78 ±\pm 0.100.88 ±\pm 0.080.83 ±\pm 0.070.74 ±\pm 0.080.73 ±\pm 0.180.75 ±\pm 0.130.76 ±\pm 0.10
Table 2: Performance comparison across external attention types, vessel categories, and smoothing levels. Values are mean ±\pm standard deviation.
Figure 3: AUC-ROC vs. σ\sigma for vessel tortuosity ( A ) and density ( B ) heatmap-attentions are shown. GradCAM results are shown in C .

Explore similar work

CardsList

  1. Deep Learning-assisted AMD Staging based on OCT and OCT Angiography

    Jun 3, 2026Yukun Guo, Tristan T. Hormel, An-Lun Wu +4Medical ImagingOptical Coherence Tomography

    Casey Eye Institute, Oregon Health & Science University, Portland, OR 97239, USA · Department of Biomedical Engineering, Oregon Health & Science University, Portland, OR 97239, USA · Department of Ophthalmology, Mackay Memorial Hospital, Hsinchu 300044, Taiwan

  2. Automated 2D and 3D Segmentation of AMD and DME Lesions in OCT

    Aug 27, 2026Lucia Sundberg, Zhihao Zhao, M. Ali NasseriMedical Image Segmentation3D Medical Image Segmentation

    MAPS Lab, Technical University of Munich, Munich, Germany. · University of Wisconsin-Madison, Madison, Wisconsin, USA.

  3. Retinal OCTA Phenotyping with LLM Reporting for Alzheimer's Disease

    Sep 4, 2026Progga Paromita Dutta, Jeba Maliha, Md Rafiul KabirExplainable Medical Image AnalysisOptical Coherence Tomography

    Department of Computer Science, Columbia University, New York, NY, USA · School of Engineering and Technology, Central Michigan University, Mount Pleasant, MI, USA