[Paper Review] A Lightweight Brain-Inspired Machine Learning Framework for Coronary Angiography: Hybrid Neural Representation and Robust Learning Strategies

Jingsong Xia, Siqi Wang|arXiv (Cornell University)|Jan 22, 2026

Retinal Imaging and Analysis0 citations

TL;DR

A lightweight brain-inspired framework for coronary angiography classification uses a pretrained ResNet50 backbone with a compact classifier, brain-inspired loss, and selective plasticity training to achieve strong performance under limited resources (Test Acc 85.00%, AUC 0.9372, high sensitivity).

ABSTRACT

Background: Coronary angiography (CAG) is a cornerstone imaging modality for assessing coronary artery disease and guiding interventional treatment decisions. However, in real-world clinical settings, angiographic images are often characterized by complex lesion morphology, severe class imbalance, label uncertainty, and limited computational resources, posing substantial challenges to conventional deep learning approaches in terms of robustness and generalization.Methods: The proposed framework is built upon a pretrained convolutional neural network to construct a lightweight hybrid neural representation. A selective neural plasticity training strategy is introduced to enable efficient parameter adaptation. Furthermore, a brain-inspired attention-modulated loss function, combining Focal Loss with label smoothing, is employed to enhance sensitivity to hard samples and uncertain annotations. Class-imbalance-aware sampling and cosine annealing with warm restarts are adopted to mimic rhythmic regulation and attention allocation mechanisms observed in biological neural systems.Results: Experimental results demonstrate that the proposed lightweight brain-inspired model achieves strong and stable performance in binary coronary angiography classification, yielding competitive accuracy, recall, F1-score, and AUC metrics while maintaining high computational efficiency.Conclusion: This study validates the effectiveness of brain-inspired learning mechanisms in lightweight medical image analysis and provides a biologically plausible and deployable solution for intelligent clinical decision support under limited computational resources.

Motivation & Objective

• Address robustness, generalization, and deployability of ML for coronary angiography under limited computational resources.
• Handle complex lesion morphology, class imbalance, and label uncertainty in real-world data.
• Develop a parameter-efficient framework leveraging neuro-inspired learning for medical imaging.
• Provide an interpretable, deployable solution for clinical decision support.

Proposed method

• Use a pretrained ResNet50 as a stable perceptual pathway with a compact linear head for classification.
• Freeze CNN backbone during initial training; train only the linear head to reduce parameter search space.
• Introduce brain-inspired attention-modulated loss by combining Focal Loss with label smoothing to address hard samples and uncertain annotations.
• Apply class-imbalance-aware sampling and cosine annealing with warm restarts to mimic rhythmic regulation and attention allocation.
• Adopt a Selective Neural Plasticity Training Strategy with staged updates: warmup with frozen backbone, then selective fine-tuning of high-level layers.
• Achieve a lightweight design with total trainable parameters around 11.4M and rapid training (≈2.2 minutes on CPU).

Experimental results

Research questions

• RQ1 Can a lightweight, brain-inspired framework achieve robust coronary angiography classification under constrained computational resources?
• RQ2 Does selective neural plasticity ( staged layer freezing and fine-tuning) improve stability and generalization in limited-data medical imaging tasks?
• RQ3 Do brain-inspired loss and uncertainty modeling (Focal Loss with label smoothing) enhance performance under class imbalance and noisy annotations?
• RQ4 How does the proposed method compare to baseline CNNs and quantum-enhanced models in accuracy, AUC, and F1-score on CAG data?

Key findings

• On a test set of 120 CAG images (60 positive, 60 negative), the model achieved 85.00% accuracy and 0.9372 AUC-ROC.
• Recall (sensitivity) was 96.67% (58/60), specificity 73.33% (44/60), and F1-score 0.8657.
• The model converged to 90% validation accuracy in 4 epochs, with validation accuracy 90.08% and AUC 0.9374 at epoch 4.
• Compared to baselines (ResNet18, ResNet34, and Quantum-Enhanced ResNet18), the brain-inspired hybrid achieved higher accuracy (85.00% vs 76.67–79.17%), higher AUC (0.9372 vs 0.8453–0.8721), and higher F1-score (0.8657 vs 0.7821–0.8234).
• Training time to target performance was about 2.2 minutes on CPU, with each epoch ~33 seconds.