Self-Supervised Learning for Improved Optical Coherence Tomography Detection of Macular Telangiectasia Type 2

Importance

Deep learning image analysis often depends on large, labeled datasets, which are difficult to obtain for rare diseases.

Objective

To develop a self-supervised approach for automated classification of macular telangiectasia type 2 (MacTel) on optical coherence tomography (OCT) with limited labeled data.

Design, Setting, and Participants

This was a retrospective comparative study. OCT images from May 2014 to May 2019 were collected by the Lowy Medical Research Institute, La Jolla, California, and the University of Washington, Seattle, from January 2016 to October 2022. Clinical diagnoses of patients with and without MacTel were confirmed by retina specialists. Data were analyzed from January to September 2023.

Exposures

Two convolutional neural networks were pretrained using the Bootstrap Your Own Latent algorithm on unlabeled training data and fine-tuned with labeled training data to predict MacTel (self-supervised method). ResNet18 and ResNet50 models were also trained using all labeled data (supervised method).

Main Outcomes and Measures

The ground truth yes vs no MacTel diagnosis is determined by retinal specialists based on spectral-domain OCT. The models' predictions were compared against human graders using accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), area under precision recall curve (AUPRC), and area under the receiver operating characteristic curve (AUROC). Uniform manifold approximation and projection was performed for dimension reduction and GradCAM visualizations for supervised and self-supervised methods.

Results

A total of 2636 OCT scans from 780 patients with MacTel and 131 patients without MacTel were included from the MacTel Project (mean [SD] age, 60.8 [11.7] years; 63.8% female), and another 2564 from 1769 patients without MacTel from the University of Washington (mean [SD] age, 61.2 [18.1] years; 53.4% female). The self-supervised approach fine-tuned on 100% of the labeled training data with ResNet50 as the feature extractor performed the best, achieving an AUPRC of 0.971 (95% CI, 0.969-0.972), an AUROC of 0.970 (95% CI, 0.970-0.973), accuracy of 0.898%, sensitivity of 0.898, specificity of 0.949, PPV of 0.935, and NPV of 0.919. With only 419 OCT volumes (185 MacTel patients in 10% of labeled training dataset), the ResNet18 self-supervised model achieved comparable performance, with an AUPRC of 0.958 (95% CI, 0.957-0.960), an AUROC of 0.966 (95% CI, 0.964-0.967), and accuracy, sensitivity, specificity, PPV, and NPV of 90.2%, 0.884, 0.916, 0.896, and 0.906, respectively. The self-supervised models showed better agreement with the more experienced human expert graders.

Conclusions and Relevance

The findings suggest that self-supervised learning may improve the accuracy of automated MacTel vs non-MacTel binary classification on OCT with limited labeled training data, and these approaches may be applicable to other rare diseases, although further research is warranted.

Developing a Continuous Severity Scale for Macular Telangiectasia Type 2 Using Deep Learning and Implications for Disease Grading

PURPOSE

Deep learning (DL) models have achieved state-of-the-art medical diagnosis classification accuracy. Current models are limited by discrete diagnosis labels, but could yield more information with diagnosis in a continuous scale. We developed a novel continuous severity scaling system for macular telangiectasia (MacTel) type 2 by combining a DL classification model with uniform manifold approximation and projection (UMAP).

DESIGN

We used a DL network to learn a feature representation of MacTel severity from discrete severity labels and applied UMAP to embed this feature representation into 2 dimensions, thereby creating a continuous MacTel severity scale.

PARTICIPANTS

A total of 2003 OCT volumes were analyzed from 1089 MacTel Project participants.

METHODS

We trained a multiview DL classifier using multiple B-scans from OCT volumes to learn a previously published discrete 7-step MacTel severity scale. The classifiers' last feature layer was extracted as input for UMAP, which embedded these features into a continuous 2-dimensional manifold. The DL classifier was assessed in terms of test accuracy. Rank correlation for the continuous UMAP scale against the previously published scale was calculated. Additionally, the UMAP scale was assessed in the κ agreement against 5 clinical experts on 100 pairs of patient volumes. For each pair of patient volumes, clinical experts were asked to select the volume with more severe MacTel disease and to compare them against the UMAP scale.

MAIN OUTCOME MEASURES

Classification accuracy for the DL classifier and κ agreement versus clinical experts for UMAP.

RESULTS

The multiview DL classifier achieved top 1 accuracy of 63.3% (186/294) on held-out test OCT volumes. The UMAP metric showed a clear continuous gradation of MacTel severity with a Spearman rank correlation of 0.84 with the previously published scale. Furthermore, the continuous UMAP metric achieved κ agreements of 0.56 to 0.63 with 5 clinical experts, which was comparable with interobserver κ values.

CONCLUSIONS

Our UMAP embedding generated a continuous MacTel severity scale, without requiring continuous training labels. This technique can be applied to other diseases and may lead to more accurate diagnosis, improved understanding of disease progression, and key imaging features for pathologic characteristics.

FINANCIAL DISCLOSURE(S)

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Author Correction: Identification of genetic factors influencing metabolic dysregulation and retinal support for MacTel, a retinal disorder

A Correction to this paper has been published: https://doi.org/10.1038/s42003-021-01972-y

Identification of genetic factors influencing metabolic dysregulation and retinal support for MacTel, a retinal disorder

Macular Telangiectasia Type 2 (MacTel) is a rare degenerative retinal disease with complex genetic architecture. We performed a genome-wide association study on 1,067 MacTel patients and 3,799 controls, which identified eight novel genome-wide significant loci (p < 5 × 10^-8), and confirmed all three previously reported loci. Using MAGMA, eQTL and transcriptome-wide association analysis, we prioritised 48 genes implicated in serine-glycine biosynthesis, metabolite transport, and retinal vasculature and thickness. Mendelian randomization indicated a likely causative role of serine (FDR = 3.9 × 10^-47) and glycine depletion (FDR = 0.006) as well as alanine abundance (FDR = 0.009). Polygenic risk scoring achieved an accuracy of 0.74 and was associated in UKBiobank with retinal damage (p = 0.009). This represents the largest genetic study on MacTel to date and further highlights genetically-induced systemic and tissue-specific metabolic dysregulation in MacTel patients, which impinges on retinal health.

Serine biosynthesis defect due to haploinsufficiency of PHGDH causes retinal disease

Macular telangiectasia type 2 (MacTel) is a progressive, late-onset retinal degenerative disease linked to decreased serum levels of serine that elevate circulating levels of a toxic ceramide species, deoxysphingolipids (deoxySLs); however, causal genetic variants that reduce serine levels in patients have not been identified. Here we identify rare, functional variants in the gene encoding the rate-limiting serine biosynthetic enzyme, phosphoglycerate dehydrogenase (PHGDH), as the single locus accounting for a significant fraction of MacTel. Under a dominant collapsing analysis model of a genome-wide enrichment analysis of rare variants predicted to impact protein function in 793 MacTel cases and 17,610 matched controls, the PHGDH gene achieves genome-wide significance (P = 1.2 × 10^-13) with variants explaining ~3.2% of affected individuals. We further show that the resulting functional defects in PHGDH cause decreased serine biosynthesis and accumulation of deoxySLs in retinal pigmented epithelial cells. PHGDH is a significant locus for MacTel that explains the typical disease phenotype and suggests a number of potential treatment options.

Serine and Lipid Metabolism in Macular Disease and Peripheral Neuropathy

BACKGROUND

Identifying mechanisms of diseases with complex inheritance patterns, such as macular telangiectasia type 2, is challenging. A link between macular telangiectasia type 2 and altered serine metabolism has been established previously.

METHODS

Through exome sequence analysis of a patient with macular telangiectasia type 2 and his family members, we identified a variant in SPTLC1 encoding a subunit of serine palmitoyltransferase (SPT). Because mutations affecting SPT are known to cause hereditary sensory and autonomic neuropathy type 1 (HSAN1), we examined 10 additional persons with HSAN1 for ophthalmologic disease. We assayed serum amino acid and sphingoid base levels, including levels of deoxysphingolipids, in patients who had macular telangiectasia type 2 but did not have HSAN1 or pathogenic variants affecting SPT. We characterized mice with low serine levels and tested the effects of deoxysphingolipids on human retinal organoids.

RESULTS

Two variants known to cause HSAN1 were identified as causal for macular telangiectasia type 2: of 11 patients with HSAN1, 9 also had macular telangiectasia type 2. Circulating deoxysphingolipid levels were 84.2% higher among 125 patients with macular telangiectasia type 2 who did not have pathogenic variants affecting SPT than among 94 unaffected controls. Deoxysphingolipid levels were negatively correlated with serine levels, which were 20.6% lower than among controls. Reduction of serine levels in mice led to increases in levels of retinal deoxysphingolipids and compromised visual function. Deoxysphingolipids caused photoreceptor-cell death in retinal organoids, but not in the presence of regulators of lipid metabolism.

CONCLUSIONS

Elevated levels of atypical deoxysphingolipids, caused by variant SPTLC1 or SPTLC2 or by low serine levels, were risk factors for macular telangiectasia type 2, as well as for peripheral neuropathy. (Funded by the Lowy Medical Research Institute and others.).

Retinal vascular permeability suppression by topical application of a novel VEGFR2/Src kinase inhibitor in mice and rabbits

Retinal and choroidal vascular diseases, with their associated abnormalities in vascular permeability, account for the majority of patients with vision loss in industrialized nations. VEGF is upregulated in ischemic retinopathies such as diabetes and is known to dramatically alter vascular permeability in a number of nonocular tissues via Src kinase-regulated signaling pathways. VEGF antagonists are currently in clinical use for treating the new blood vessels and retinal edema associated with neovascular eye diseases, but such therapies require repeated intraocular injections. We have found that vascular leakage following intravitreal administration of VEGF in mice was abolished by systemic or topical delivery of what we believe is a novel VEGFR2/Src kinase inhibitor; this was confirmed in rabbits. The relevance of Src inhibition to VEGF-associated alterations in vascular permeability was further substantiated by genetic studies in which VEGF injection or laser-induced vascular permeability failed to augment retinal vascular permeability in Src-/- and Yes-/- mice (Src and Yes are ubiquitously expressed Src kinase family members; Src-/- and Yes-/- mice lacking expression of these kinases show no vascular leak in response to VEGF). These findings establish a role for Src kinase in VEGF-mediated retinal vascular permeability and establish a potentially safe and painless topically applied therapeutic option for treating vision loss due to neovascular-associated retinal edema.

Combination angiostatic therapy completely inhibits ocular and tumor angiogenesis

Angiostatic therapies designed to inhibit neovascularization associated with multiple pathological conditions have only been partially successful; complete inhibition has not been achieved. We demonstrate synergistic effects of combining angiostatic molecules that target distinct aspects of the angiogenic process, resulting in the complete inhibition of neovascular growth associated with development, ischemic retinopathy, and tumor growth, with little or no effect on normal, mature tissue vasculature. Tumor vascular obliteration using combination angiostatic therapy was associated with reduced tumor mass and increased survival in a rat 9L gliosarcoma model, whereas individual monotherapies were ineffective. Significant compensatory up-regulation of several proangiogenic factors was observed after treatment with a single angiostatic agent. In contrast, treatment with combination angiostatic therapy significantly reduced compensatory up-regulation. Therapies that combine angiostatic molecules targeting multiple, distinct aspects of the angiogenic process may represent a previously uncharacterized paradigm for the treatment of many devastating diseases with associated pathological neovascularization.

Three-Dimensional In Vivo Imaging of the Mouse Intraocular Vasculature during Development and Disease

PURPOSE

Aberrant growth of blood vessels in the eye is a major cause of vision loss, occurring as a complication of diabetic retinopathy, age-related macular degeneration, and retinal vascular occlusions, among others. Whereas in humans, in vivo angiography is routinely used to image such diseases, many animal models of ocular vascular disease and development rely on dissected tissues that may not accurately represent in vivo conditions and require enucleation of the eye, the death of the animal, or both.

METHODS

A method of three-dimensional imaging of blood vessels was used in the living mouse eye that involved scanning laser confocal microscopy and computer-aided image reconstruction.

RESULTS

This minimally invasive technique was used to collect three-dimensional images of intraocular vessels in vivo during development. The retinal and choroidal vasculature was studied during development and disease, in models of retinal degeneration, central retinal vein occlusion, and oxygen-induced retinopathy. To aid in investigations into cell-based therapies for retinal disease, two-color imaging was used to localize transplanted cells in relation to the vasculature. This technique was used to perform serial imaging of the ocular vasculature over time, when developmental regression of vessels was observed.

CONCLUSIONS

This in vivo vascular imaging approach is valuable in monitoring normal development, disease progression, and efficacy of experimental treatments in mouse models of ocular vascular disease and may have broader applications to the field of angiogenesis by using the readily visualized ocular vascular bed as a surrogate to test pro- and antiangiogenic compounds.