DP-SENet: A Dual-Path Attention Raman Spectroscopy Analysis Network for IgA Nephropathy Diagnosis
Year: 2025
Type: conference-paper
Abstract: IgA nephropathy(IgAN) is considered to be the most common primary glomerular disease in the world.Early screening and timely intervention of IgA nephropathy are of great significance for delaying the progression of end-stage renal disease and reducing the related mortality.Because IgAN needs to be combined with the history of related diseases, clinical manifestations and laboratory examination before diagnosis, the scope of differential diagnosis is wide and complex, and invasive sampling methods may also cause complications.Therefore, it is urgent to develop a non-invasive, sensitive and efficient IgAN auxiliary diagnosis technology.This study proposes a Dual-Path Attention Network (DP-SENet) to achieve rapid and automated diagnosis of IgAN.Serum Raman spectra are acquired from a cohort of 58 IgAN patients and 41 healthy controls.To enhance the model's generalizability,data augmentation technology is used to expand the sample size.The raw spectral data undergoes preprocessing, which includes baseline correction via the adaptive iteratively reweighted penalized least squares (airPLS) algorithm and noise reduction using the Savitzky-Golay (SG) smoothing method.The DP-SENet architecture incorporates a dual-path design that synergistically integrates Convolutional Neural Networks (CNNs) with attention mechanisms for feature extraction.The upper path is equipped with a Scaled Dot-Product Attention (SDPA) module to capture long-range dependencies within the spectral sequences.The lower path uses efficient channel attention(ECA) to distinguish important features by weighting.The upper path and lower path cooperate with each other to improve classification accuracy.Experimental results demonstrate that the proposed DP-SENet achieves superior performance across multiple classification metrics compared to established deep learning models.This approach provides a novel solution for the non-invasive and efficient diagnosis of IgAN, holding significant clinical reference value and potential for application in enhancing the precision of auxiliary renal disease diagnosis.
Language: English
Subfield: Biophysics
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7124162005