Neurovascular Retinomics for Noninvasive Profiling of Cancer Risk
8 affiliations
- 1School of Optometry, The Hong Kong Polytechnic University, Kowloon, Hong Kong.
- 2Research Centre for SHARP Vision, The Hong Kong Polytechnic University, Kowloon, Hong Kong.
- 3Centre for Eye Research Australia, Royal Victorian Eye and Ear Hospital, East Melbourne, Australia.
- 4Department of Surgery (Ophthalmology), The University of Melbourne, Melbourne, Australia.
- 5Department of Ophthalmology, Guangdong Eye Institute, Guangdong Provincial People's Hospital, Guangdong Academy of Medical Sciences, Guangzhou, China.
- 6Department of Colorectal Surgery, The Sixth Affiliated Hospital, Sun Yat-sen University, Guangzhou, China.
- 7InnoHK, Centre for Eye and Vision Research, Kowloon, Hong Kong.
- 8PolyU-Stanford Joint Research Centre for Longitudinal Deep Omics, The Hong Kong Polytechnic University, Kowloon, Hong Kong.
Abstract
Purpose: To comprehensively evaluate the predictive value of neurovascular retinal measurements for incident cancer.
Design: Population-based, prospective cohort study.
Participants: A total of 41 424 participants with eligible color fundus photography and OCT data from the UK Biobank.
Methods: Retinomics were derived from multimodal retinal imaging of the UK Biobank, comprising 156 retinal features. Four Cox regression models were developed: model 1 (age and sex), model 2 (model 1 + ethnicity and lifestyle factors), model 3 (model 1 + retinomics), and model 4 (full model). Least Absolute Shrinkage and Selection Operator regression was used to select relevant retinal biomarkers, tailored to each cancer type. Predictive performance was assessed via the concordance index (C-index), detection rate (DR), and positive likelihood ratio (LR+). Detection rate and LR+ were evaluated at a false-positive rate of 10%. Calibration was assessed by calibration slope and calibration-in-the-large (CITL). Hazard ratios (HRs) were used to stratify cancer risk between high- and low-risk groups identified by retinomics.
Main Outcome Measures: Incident diagnoses of any cancer and specific types: breast, uterine/endometrial, prostate, esophageal, lung, stomach, colon, rectal, skin, and other cancers.
Results: Among 41 424 participants, 3633 (8.77%) developed cancer over 12.5 years of follow-up. Incorporating retinomics significantly improved the C-index, DR, and LR+ across most cancer types, with the C-index values for model 4 ranging from 0.581 to 0.885 across individual cancers. The largest gain was observed in uterine/endometrial cancer (ΔC-index = 0.220; 95% confidence interval [CI]: 0.119–0.327]; ΔDR = 40%; ΔLR+ = 4.003), although this finding is based on only 20 incident cases and should be interpreted with caution. Calibration analysis demonstrated good agreement between predicted and observed risks across all models (slope range: 0.635–1.233; CITL near 0). The high-risk group identified by retinomics exhibited a 7% higher annual cancer risk (HR = 1.07; 95% CI: 1.06–1.08; P < 0.001), accumulating to a nearly threefold higher risk over 12 years.
Conclusions: Retinomics demonstrates predictive value for cancer incidence, with cancer type–specific biomarker selection enhancing accuracy. These findings highlight the transformative potential of retinal imaging as a scalable, noninvasive approach for early cancer risk profiling and proactive prevention.
Financial Disclosure(s): Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.
Related
- Longitudinal ctDNA profiling reveals genomic risk and molecular response to bireociclib in advanced breast cancer
- Identifying a high-risk CKM syndrome phenotype: association of albuminuria with subclinical hemodynamic alterations via portable noninvasive cardiometer
- Risk factor analysis and development of a risk prediction model for facial pressure injury associated with noninvasive positive pressure ventilation in patients with chronic obstructive pulmonary disease based on symptom science theory
- IoT-enabled multimodal large language model-based framework for compositional food recognition, nutrient profiling, NOVA classification, and cancer risk mitigation
- Analytical and clinical validation of the EdenDx methylation assay as a noninvasive supportive triage tool for early detection of endometrial cancer in cervical scraping specimens
BibTeX
@article{zhang2026neurovascular,
title = {{Neurovascular Retinomics for Noninvasive Profiling of Cancer Risk}},
author = {Selena Wei Zhang and Mayinuer Yusufu and Xia Gong and Xianwen Shang and Feng Gao and Mingguang He and Danli Shi},
journal = {Ophthalmology Science},
year = {2026},
month = sep,
doi = {10.1016/j.xops.2026.101413},
url = {https://doi.org/10.1016/j.xops.2026.101413},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Neurovascular Retinomics for Noninvasive Profiling of Cancer Risk
AU - Selena Wei Zhang
AU - Mayinuer Yusufu
AU - Xia Gong
AU - Xianwen Shang
AU - Feng Gao
AU - Mingguang He
AU - Danli Shi
PY - 2026
DA - 2026/09/21
JO - Ophthalmology Science
DO - 10.1016/j.xops.2026.101413
UR - https://doi.org/10.1016/j.xops.2026.101413
AB - Purpose: To comprehensively evaluate the predictive value of neurovascular retinal measurements for incident cancer. Design: Population-based, prospective cohort study. Participants: A total of 41 424 participants with eligible color fundus photography and OCT data from the UK Biobank. Methods: Retinomics were derived from multimodal retinal imaging of the UK Biobank, comprising 156 retinal features. Four Cox regression models were developed: model 1 (age and sex), model 2 (model 1 + ethnicity and lifestyle factors), model 3 (model 1 + retinomics), and model 4 (full model). Least Absolute Shrinkage and Selection Operator regression was used to select relevant retinal biomarkers, tailored to each cancer type. Predictive performance was assessed via the concordance index (C-index), detection rate (DR), and positive likelihood ratio (LR+). Detection rate and LR+ were evaluated at a false-positive rate of 10%. Calibration was assessed by calibration slope and calibration-in-the-large (CITL). Hazard ratios (HRs) were used to stratify cancer risk between high- and low-risk groups identified by retinomics. Main Outcome Measures: Incident diagnoses of any cancer and specific types: breast, uterine/endometrial, prostate, esophageal, lung, stomach, colon, rectal, skin, and other cancers. Results: Among 41 424 participants, 3633 (8.77%) developed cancer over 12.5 years of follow-up. Incorporating retinomics significantly improved the C-index, DR, and LR+ across most cancer types, with the C-index values for model 4 ranging from 0.581 to 0.885 across individual cancers. The largest gain was observed in uterine/endometrial cancer (ΔC-index = 0.220; 95% confidence interval [CI]: 0.119–0.327]; ΔDR = 40%; ΔLR+ = 4.003), although this finding is based on only 20 incident cases and should be interpreted with caution. Calibration analysis demonstrated good agreement between predicted and observed risks across all models (slope range: 0.635–1.233; CITL near 0). The high-risk group identified by retinomics exhibited a 7% higher annual cancer risk (HR = 1.07; 95% CI: 1.06–1.08; P < 0.001), accumulating to a nearly threefold higher risk over 12 years. Conclusions: Retinomics demonstrates predictive value for cancer incidence, with cancer type–specific biomarker selection enhancing accuracy. These findings highlight the transformative potential of retinal imaging as a scalable, noninvasive approach for early cancer risk profiling and proactive prevention. Financial Disclosure(s): Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.
SN - 2666-9145
ER -