Deep Learning–Driven Surface‐Enhanced Raman Scattering–Lateral Flow Immunoassay With Au Nanostars for High‐Accuracy Respiratory Virus Detection

ABSTRACT The high accuracy in surface‐enhanced Raman scattering‐lateral flow immunoassays (SERS–LFIAs) is critical for reliable point‐of‐care testing (POCT) in clinical diagnostics. Conventional approaches are often affected by sampling variability and uneven distribution of immunoprobes, leading to unreliable signal fluctuations. To address this challenge, we developed a high‐performance SERS–LFIA strip based on gold nanostars (Au NSs) and integrated it with an artificial intelligence (AI)–powered diagnostic framework. Specifically, Au NSs with exceptional SERS enhancement were synthesized via an optimized “two‐step” method and utilized as nanoprobes to construct an influenza B (FluB) SERS–LFIA strip for performance validation. A novel large‐area Raman scanning technique was then employed to generate intensity maps depicting the immunoprobe distribution around the test ( T ) line. A deep residual neural network (ResNet‐18) was subsequently applied to analyze these SERS images, minimizing subjective interpretation and significantly improving accuracy. The optimized framework achieved 100% training accuracy and 95% validation accuracy, significantly outperforming conventional peak intensity analysis and support vector machine (SVM)–based full‐spectrum discrimination methods. The Au NSs–based SERS–LFIA platform and the optimized ResNet‐18 model were integrated into a portable Raman spectrometer to create an automated diagnostic system. To further evaluate the stability and versatility of the system, the detection target was switched to influenza A (FluA) by altering the capture and detection antibodies. This reengineered system demonstrated a 95% accuracy rate in testing 40 simulated human clinical samples. Our work establishes a machine learning–enhanced, automated SERS–LFIA system that leverages Au NSs for superior signal enhancement and utilizes deep learning for robust image analysis. This integrated approach provides a scalable and high‐performance POCT framework, paving the way for automated clinical diagnostics.