Truace
Subscribe

Deep learning detection of ocular toxoplasmosis from fundus photography

Source article: Diagnostic Performance of Deep Learning Models for Ocular Toxoplasmosis Using Fundus Photography: A Systematic Review and Meta-Analysis

Background Ocular toxoplasmosis (OT) is the leading infectious cause of posterior uveitis worldwide, yet diagnosis remains challenging without a universal gold standard. Although deep learning (DL) shows significant promise for automated image analysis, a rigorous synthesis of its diagnostic accuracy across heterogeneous architectures and datasets is lacking. Methods Following PRISMA-DTA and CLAIM-AI guidelines, we searched major databases through January 2026 for studies evaluating DL models for OT detection us…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1770· cite & verify
Diagnostic Performance of Deep Learning Models for Ocular Toxoplasmosis Using Fundus Photography: A Systematic Review and Meta-Analysis
Hospital Universitari Doctor Peset, València 11 by 19Tarrestnom65. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0
Trace impact readingContested
P 68The P score combines the specificity and measured human impact of the grounded problem claim with the strength of this Trace’s cited sources.

Both sides are scored from claims and sources, not community votes.

G 65The G score combines the specificity and measured human impact of the grounded gain claim with the strength of this Trace’s cited sources.

In brief

A systematic review and meta-analysis searched databases through January 2026 for studies testing deep learning on fundus photographs to detect ocular toxoplasmosis. Sixteen studies were synthesized qualitatively, and nine with usable contingency data were pooled, showing sensitivity of 0.965 and specificity of 0.955.

The results suggest automated image analysis could support diagnosis of a leading infectious cause of posterior uveitis, where no universal gold standard exists. Yet the evidence base is geographically narrow and heterogeneous, so real-world clinical performance in diverse populations remains unproven.

Main points

  1. Sixteen studies covering 38 model configurations were included; nine studies with 24 architectures were pooled.
  2. The pooled diagnostic odds ratio was 594.1, with positive and negative likelihood ratios of 21.6 and 0.036.
  3. Between-study heterogeneity was substantial, with I^2 of 87.9%.
  4. 81.2% of studies relied exclusively on the single OTFID dataset from Paraguay.
  5. The authors call for diverse international cohorts, explainable AI, and federated learning.

The gain

Across nine meta-analyzed studies, deep learning models detected ocular toxoplasmosis on fundus photographs with high pooled sensitivity and specificity.

The problem

Most included studies used one Paraguayan dataset and lacked multicentric external validation, which limits how far the accuracy can be generalized to other populations.

The rundown

The review followed PRISMA-DTA and CLAIM-AI guidelines, appraised study quality with QUADAS-2, and used a bivariate random-effects model to pool sensitivity and specificity, with HSROC curves to visualize discriminatory performance.

Beyond accuracy, the authors report the pooled diagnostic odds ratio of 594.1 and note the performance is comparable to established ophthalmic AI benchmarks. They caution that the high heterogeneity complicates interpretation of the pooled figures.

The authors raise concerns about phenotypic bias from reliance on one dataset and say clinical translation is constrained by the lack of multicentric external validation. They recommend future work prioritize diverse international cohorts, explainable AI, and federated learning.

What this doesn’t fix

Most included studies used a single Paraguayan dataset, with high heterogeneity and no multicentric external validation, so global generalizability is uncertain.

Sources

  1. Peer-reviewedOcular Immunology and Inflammation2026-10-07

The debate