[Paper Review] CLEAR-Mamba:Towards Accurate, Adaptive and Trustworthy Multi-Sequence Ophthalmic Angiography Classification

Zhuonan Wang, Wenjie Yan+6 more|arXiv (Cornell University)|Jan 28, 2026

Retinal Imaging and Analysis0 citations

TL;DR

CLEAR-Mamba enhances single-modality (FFA/ICGA) ophthalmic angiography classification by incorporating hypernetwork-based adaptation and evidential reliability, achieving better temporal use, generalization, and calibrated uncertainty on a 43-disease in-house dataset.

ABSTRACT

Medical image classification is a core task in computer-aided diagnosis (CAD), playing a pivotal role in early disease detection, treatment planning, and patient prognosis assessment. In ophthalmic practice, fluorescein fundus angiography (FFA) and indocyanine green angiography (ICGA) provide hemodynamic and lesion-structural information that conventional fundus photography cannot capture. However, due to the single-modality nature, subtle lesion patterns, and significant inter-device variability, existing methods still face limitations in generalization and high-confidence prediction. To address these challenges, we propose CLEAR-Mamba, an enhanced framework built upon MedMamba with optimizations in both architecture and training strategy. Architecturally, we introduce HaC, a hypernetwork-based adaptive conditioning layer that dynamically generates parameters according to input feature distributions, thereby improving cross-domain adaptability. From a training perspective, we develop RaP, a reliability-aware prediction scheme built upon evidential uncertainty learning, which encourages the model to emphasize low-confidence samples and improves overall stability and reliability. We further construct a large-scale ophthalmic angiography dataset covering both FFA and ICGA modalities, comprising multiple retinal disease categories for model training and evaluation. Experimental results demonstrate that CLEAR-Mamba consistently outperforms multiple baseline models, including the original MedMamba, across various metrics-showing particular advantages in multi-disease classification and reliability-aware prediction. This study provides an effective solution that balances generalizability and reliability for modality-specific medical image classification tasks. Our project can be accessed at https://github.com/ZJU4HealthCare/CLEAR-Mamba.

Motivation & Objective

• Motivate robust angiography classification by leveraging temporal information in single-modality FFA/ICGA sequences.
• Improve cross-domain adaptability with case-conditioned parameter generation.
• Enhance prediction reliability through calibrated uncertainty and risk-aware deferral.
• Construct and utilize a large-scale 43-disease ophthalmic angiography dataset (FFA/ICGA) for training and evaluation.

Proposed method

• Use MedMamba as an efficient temporal backbone to capture long-range dynamics via Visual State Space Models (VSSMs).
• Introduce HaC, a HyperNetwork-based adaptive conditioning layer, to generate case-specific parameters for instance-conditioned feature modulation.
• Incorporate RaP, a reliability-aware prediction head based on evidential learning to output calibrated probabilities and uncertainty.
• Train with an evidential objective that combines negative log-likelihood with a Dirichlet prior regularizer for uncertainty estimation.
• Evaluate CLEAR-Mamba at three scales (T/S/B) and compare against CNN/ViT/Mamba baselines on in-house and public datasets.

Figure 1: Automated pipeline for extracting, anonymizing, and aligning medical image–text data from raw PDF reports.

Experimental results

Research questions

• RQ1 Can single-modality ophthalmic angiography (FFA/ICGA) be significantly improved for multi-disease classification by exploiting temporal dynamics?
• RQ2 Do hypernetwork-based adaptation (HaC) and evidential reliability (RaP) jointly improve accuracy and prediction calibration under domain shifts?
• RQ3 How does CLEAR-Mamba perform across different model scales (Tiny/Small/Big) compared to medical-specific and general architectures?
• RQ4 Does the approach provide reliable uncertainty estimates that support risk-aware clinical decision making?
• RQ5 Is a large, real-world, single-modality angiography dataset with 43 diseases viable for training robust models?

Key findings

• CLEAR-Mamba consistently outperforms baselines including MedMamba across metrics (OA, F1, AUC) on the in-house 43-disease dataset.
• HaC improves feature conditioning while RaP provides calibrated uncertainty, with the full model achieving the best OA/AUC.
• Full CLEAR achieves higher sensitivity and OA than MedMamba-X across scales (T/S/B).
• Ablations show HaC alone can be overconfident; RaP alone can be overly conservative; together they yield better calibration and discrimination.
• t-SNE embeddings indicate CLEAR produces tighter intra-class clustering and clearer inter-class separation than competing models.
• Uncertainty metrics (predictive entropy) effectively indicate when manual review is warranted, supporting risk-aware deployment.

Figure 2: Dataset statistics. (a) Class distribution across 43 ocular categories showing a long-tailed pattern. (b) Modality proportion between FFA and ICGA images.