Deep learning assisted quantitative detection of cardiac troponin I in hierarchical dendritic copper-nickel nanostructures lateral flow immunoassay

Shenglan Zhang,Liqiang Chen,YuXin Tan,Shaojie Wu,Pengxin Guo,Xincheng Jiang,Hongcheng Pan

DOI: https://doi.org/10.1039/d4ay01187b

IF: 3.1

2024-09-04

Analytical Methods

Abstract:The rising demand for point-of-care testing (POCT) in disease diagnosis has made LFIA sensors based on dendritic metal thin film (HD-nanoMetal) and background fluorescence technology essential for rapid and accurate disease marker detection, thanks to their integrated design, high sensitivity, and cost-effectiveness. However, their unique 3D nanostructures cause significant fluorescence variation, challenging traditional image processing methods in segmenting weak fluorescence regions. This paper develops a deep learning method to efficiently segment target regions in HD-nanoMetal LFIA sensor images, improving quantitative detection accuracy. We propose an improved Unet++ network with attention and residual modules, accurately segmenting varying fluorescence intensities, especially weak ones. We evaluated the method using IoU and Dice coefficients, comparing it with Unet, Deeplabv3, and Unet++. We used an HD-NanoCu-Ni LFIA sensor for cardiac troponin I (cTnI) as a case study to validate the method's practicality. The proposed method achieved a 96.3% IoU, outperforming other networks. The R2 between characteristic quantity and cTnI concentration reached 0.994, confirming the method's accuracy and reliability. This enhances POCT accuracy and provides a reference for future fluorescence immunochromatography expansion.

spectroscopy,chemistry, analytical,food science & technology