AI‐assisted automated endpoint interpretation for turbidity‐based digital LAMP using bright‐field microscopy

Digital loop‐mediated isothermal amplification based on turbidity (dLAMP) is a nucleic acid detection technique that combines high sensitivity and specificity with low cost, offering substantial potential for deployment in resource‐limited settings. However, endpoint interpretation relying on bright‐field microscopic imaging commonly suffers from insufficient contrast, blurred target boundaries, and pronounced noise interference, which severely hampers automated and large‐scale identification of positive reaction units. To address these challenges, we propose NanoFuse‐YOLO11, an object detection model tailored for weak‐contrast microscopic turbidity images. The model incorporates a dynamic feature representation module to enhance structural perception in low‐contrast regions, while integrating an adaptive multi‐scale spectral attention module to effectively suppress background noise and emphasize critical diagnostic features. In addition, the WIoUv3 regression loss is employed to improve localization accuracy and stability under high intersection over union thresholds. Experimental results on a self‐constructed turbidity‐based dLAMP dataset demonstrate that the proposed method achieves 99.3% mAP@0.5 and 89.6% mAP@0.5:0.95, substantially outperforming You Only Look Once (YOLO) variants and several mainstream object detection models, while maintaining efficient inference performance with controlled model size and computational overhead. Overall, NanoFuse‐YOLO11 provides an effective automated approach for endpoint interpretation of bright‐field turbidity‐based dLAMP. The proposed method may facilitate fluorescence‐free digital nucleic acid quantification and has potential applicability to other weak‐contrast microscopic imaging detection tasks.