Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Discriminating retinal microvascular and neuronal differences related to migraines: Deep Learning based Crossectional Study

Feilong Tang, Matt Trinh, Annita Duong, Angelica Ly, Fiona Stapleton, Zhe Chen, Zongyuan Ge, Imran Razzak

TL;DR

This study investigates whether retinal imaging can distinguish individuals with migraines from controls using deep learning on color fundus photography (CFP) and optical coherence tomography (OCT). By evaluating three CNN backbones (VGG-16, ResNet-50, Inceptionv3) on two CFP data types (posterior pole and optic nerve head) with and without OCT-derived ONH measurements, the authors find that wide-field CFP data (posterior pole) achieves strong discrimination (AUC ~0.84–0.87), while ONH-focused CFP data performs substantially worse. OCT measurements provide modest, non-significant improvements, and class activation maps consistently emphasize the paravascular arcades, suggesting retinal microvascular differences are more informative than neuronal RNFL thickness alone. Overall, the work supports retinal vasculature as a potential non-invasive biomarker for migraine and highlights the value of broad retinal imaging, while noting limitations such as sample size and reliance on self-reported diagnoses. Further work should expand datasets, incorporate additional imaging modalities, and pursue interpretable models across diverse retinal conditions.

Abstract

Migraine, a prevalent neurological disorder, has been associated with various ocular manifestations suggestive of neuronal and microvascular deficits. However, there is limited understanding of the extent to which retinal imaging may discriminate between individuals with migraines versus without migraines. In this study, we apply convolutional neural networks to color fundus photography (CFP) and optical coherence tomography (OCT) data to investigate differences in the retina that may not be apparent through traditional human-based interpretations of retinal imaging. Retrospective data of CFP type 1 [posterior pole] and type 2 [optic nerve head (ONH)] from 369 and 336 participants respectively were analyzed. All participants had bilaterally normal optic nerves and maculae, with no retinal-involving diseases. CFP images were concatenated with OCT default ONH measurements, then inputted through three convolutional neural networks - VGG-16, ResNet-50, and Inceptionv3. The primary outcome was performance of discriminating between with migraines versus without migraines, using retinal microvascular and neuronal imaging characteristics. Using CFP type 1 data, discrimination (AUC [95% CI]) was high (0.84 [0.8, 0.88] to 0.87 [0.84, 0.91]) and not significantly different between VGG-16, ResNet-50, and Inceptionv3. Using CFP type 2 [ONH] data, discrimination was reduced and ranged from poor to fair (0.69 [0.62, 0.77] to 0.74 [0.67, 0.81]). OCT default ONH measurements overall did not significantly contribute to model performance. Class activation maps (CAMs) highlighted that the paravascular arcades were regions of interest. The findings suggest that individuals with migraines demonstrate microvascular differences more so than neuronal differences in comparison to individuals without migraines.

Discriminating retinal microvascular and neuronal differences related to migraines: Deep Learning based Crossectional Study

TL;DR

This study investigates whether retinal imaging can distinguish individuals with migraines from controls using deep learning on color fundus photography (CFP) and optical coherence tomography (OCT). By evaluating three CNN backbones (VGG-16, ResNet-50, Inceptionv3) on two CFP data types (posterior pole and optic nerve head) with and without OCT-derived ONH measurements, the authors find that wide-field CFP data (posterior pole) achieves strong discrimination (AUC ~0.84–0.87), while ONH-focused CFP data performs substantially worse. OCT measurements provide modest, non-significant improvements, and class activation maps consistently emphasize the paravascular arcades, suggesting retinal microvascular differences are more informative than neuronal RNFL thickness alone. Overall, the work supports retinal vasculature as a potential non-invasive biomarker for migraine and highlights the value of broad retinal imaging, while noting limitations such as sample size and reliance on self-reported diagnoses. Further work should expand datasets, incorporate additional imaging modalities, and pursue interpretable models across diverse retinal conditions.

Abstract

Migraine, a prevalent neurological disorder, has been associated with various ocular manifestations suggestive of neuronal and microvascular deficits. However, there is limited understanding of the extent to which retinal imaging may discriminate between individuals with migraines versus without migraines. In this study, we apply convolutional neural networks to color fundus photography (CFP) and optical coherence tomography (OCT) data to investigate differences in the retina that may not be apparent through traditional human-based interpretations of retinal imaging. Retrospective data of CFP type 1 [posterior pole] and type 2 [optic nerve head (ONH)] from 369 and 336 participants respectively were analyzed. All participants had bilaterally normal optic nerves and maculae, with no retinal-involving diseases. CFP images were concatenated with OCT default ONH measurements, then inputted through three convolutional neural networks - VGG-16, ResNet-50, and Inceptionv3. The primary outcome was performance of discriminating between with migraines versus without migraines, using retinal microvascular and neuronal imaging characteristics. Using CFP type 1 data, discrimination (AUC [95% CI]) was high (0.84 [0.8, 0.88] to 0.87 [0.84, 0.91]) and not significantly different between VGG-16, ResNet-50, and Inceptionv3. Using CFP type 2 [ONH] data, discrimination was reduced and ranged from poor to fair (0.69 [0.62, 0.77] to 0.74 [0.67, 0.81]). OCT default ONH measurements overall did not significantly contribute to model performance. Class activation maps (CAMs) highlighted that the paravascular arcades were regions of interest. The findings suggest that individuals with migraines demonstrate microvascular differences more so than neuronal differences in comparison to individuals without migraines.
Paper Structure (21 sections, 3 equations, 4 figures, 8 tables)

This paper contains 21 sections, 3 equations, 4 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Dataset collection
  4. Overview of the study methodology
  5. Experiment
  6. Result
  7. CFP Data Types
  8. Impact of OCT Measurements
  9. Statistical Significance
  10. OCT measurements
  11. Model Comparison
  12. CAMs Visualization
  13. Discussion
  14. Retinal Vascular Pathophysiological Theories
  15. Significance of Retinal Imaging
  16. ...and 6 more sections

Figures (4)

  • Figure 1: Fundus photo of right eye (migraine vs healthy); Posterior Pole : \ref{['fig:ex3-a']} Migraine; \ref{['fig:ex3-b']} Healthy; Optic Nerve Head \ref{['fig:ex3-c']} Migraine, \ref{['fig:ex3-d']} Healthy
  • Figure 2: (A) Dataset Construction: The study used data from the Centre for Eye Health. The data were derived from a practical scenario in routine clinical practice. (B) Data Processing: CFP were divided into Type 1 data (posterior pole) and Type 2 data (optic nerve head). Data was further split into training and testing sets. (C) Model Development: Features were extracted from both CFP data types and OCT measurements to discriminate between migraine and non-migraine subjects.
  • Figure 3: ROC curves for the CFP Type 1 and Type 2 data with and without ONH measurements.
  • Figure 4: Class Activation Maps (CAM) examples. CAMs generally demonstrated focus at the paravascular arcades. An example is provided for eyes/participants with (A) migraines and (B) no migraines.