Label-correlation model improves detection of multiple retinal diseases
Beyond Multi-Stage Pipelines: A Label-Correlation Ensemble for Multi-Label Retinal Fundus Disease Classification
Computer Vision and Pattern Recognition
Summary
Retinal images often show several diseases at once, making diagnosis challenging. The authors compared three different methods for spotting multiple eye diseases in one image. They found that a model that looks at relationships between diseases works better than methods that filter cases in stages or treat each disease independently. Combining this model with a standard approach improves detection even more. This helps especially with rare diseases that usually have fewer examples in the data.
What this means in practice
- •For medical imaging teams: Implement multi-label retinal disease screening systems that use label-correlation models to improve detection accuracy, especially for uncommon conditions.
- •For ophthalmology software developers: Develop diagnostic tools that combine flat and label-correlation models with test-time augmentation to enhance retinal disease classification performance.$Commercial implications: Enables creation of commercial diagnostic AI software for eye disease screening with improved accuracy and reliability.
Authors
Md Asif Uddin, Jannatul Ferdous, Asif Islam, Zobaer Mahmud Zisan, Promit Ghosh, Shah Ahafik Arman, Niloy Farhan, Rafeed Rahman
Abstract
Retinal fundus images frequently carry more than one disease, and public fundus datasets are heterogeneous and severely long-tailed. Clinical triage suggests a staged design, in which a screening gate, a disease-group router and group-specific classifiers operate in sequence, but it is unclear whether this hierarchy earns its complexity. We present a controlled comparison of three architectures on a unified 39-target label space built from RFMiD 1.0, RFMiD 2.0 and ODIR-5K (12,652 images): (i) a three-stage gated cascade of eight models, (ii) a flat model with 39 independent sigmoid outputs, and (iii) a label-correlation model that replaces the flat linear head with 39 learned label queries and a two-layer transformer decoder. All three share one Swin-V2-B backbone, one preprocessing pipeline and one training protocol, so differences are attributable to the architecture. Staging offers no consistent advantage: tracing every missed disease instance to the stage that lost it shows that the two coarse gates discard 34.2% of true instances while the fine-grained classifiers lose only 6.7%, and the gates' hard zeros depress the cascade's macro-AUC to 0.747. Label-correlation modeling gives the largest single gain, raising the detection rate from 0.621 to 0.682 (0.689 with test-time augmentation). Blending the flat and label-correlation models under four-view test-time augmentation is the best configuration on six of eight metrics (exact match 0.694, micro-F1 0.678, macro-F1 0.456, macro-AUC 0.972). On definition-matched class subsets it matches or exceeds four recent studies on macro- and micro-F1, most clearly on low-support pathologies. We also show that ensembling a gated cascade with continuous-output models is harmful, and that per-source negative sampling is needed to remove a dataset shortcut that otherwise yields a spurious AUC of 1.000.