Answers are grounded in this paper and cite their sources.
Automated malaria diagnosis using microscopic blood smear images has emerged as an effective alternative to manual examination; however, existing deep learning approaches predominantly rely on spatial-domain representations and conventional optimization objectives, limiting their ability to exploit complementary spectral information and discriminate challenging samples. This paper proposes WaveFusionNet, a novel dual-branch deep learning framework that jointly learns spatial and frequency-domain representations for robust malaria cell classification. The proposed architecture introduces two original contributions. First, a Frequency-Gated Spatial Modulation (FGSM) module computes wavelet detail-energy maps to generate adaptive spatial gates, enabling frequency information to selectively enhance informative spatial features while suppressing irrelevant background responses. Second, a Wavelet Energy-Guided Contrastive Loss (WEGCL) exploits eight-dimensional wavelet energy signatures to identify spectrally hard positive and negative pairs, thereby strengthening feature discrimination through adaptive contrastive learning. Furthermore, a Cross-Domain Attention Fusion module effectively integrates complementary spatial and frequency representations using multi-head cross-attention, while Adaptive Channel-Spatial Attention dynamically balances channel-wise and spatial feature refinement throughout the network. Extensive experiments on the benchmark NIH Malaria Cell Images dataset (27,558 images) demonstrate that WaveFusionNet achieves 98.47% accuracy, 98.60% recall, and an AUC of 0.9978, consistently outperforming eight state-of-the-art deep learning models. Five-fold cross-validation confirms excellent statistical robustness, and Grad-CAM visualizations verify that the proposed framework accurately localizes clinically relevant parasitic regions. Despite its superior predictive capability, the model remains computationally efficient, with only 15.1 million parameters and an inference latency of 7.2 ms, making it well-suited for real-time deployment in resource-constrained point-of-care environments. The proposed WaveFusionNet provides an effective and interpretable spatial-frequency learning paradigm that advances automated malaria diagnosis through synergistic feature fusion and wavelet-guided representation learning.
No takes yet. Share an insight, caveat, or question.
Alzahrani et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: