Attentional Triple-Encoder Network in Spatiospectral Domains for Medical Image Segmentation
This is an incremental improvement for medical image segmentation in ophthalmology.
The paper tackled retinal OCT segmentation by integrating spatial and spectral domains with a triple-encoder network, achieving an average Dice score improvement from 0.855 to 0.864.
Retinal Optical Coherence Tomography (OCT) segmentation is essential for diagnosing pathology. Traditional methods focus on either spatial or spectral domains, overlooking their combined dependencies. We propose a triple-encoder network that integrates CNNs for spatial features, Fast Fourier Convolution (FFC) for spectral features, and attention mechanisms to capture global relationships across both domains. Attention fusion modules integrate convolution and cross-attention to further enhance features. Our method achieves an average Dice score improvement from 0.855 to 0.864, outperforming prior work.