Precise, fast, and automated gel quantification powered by YOLO11 instance segmentation

Gel electrophoresis is a cornerstone technique in analytical chemistry and proteomics, yet downstream image analysis remains a bottleneck constrained by manual labor and subjectivity. Traditional profile-based densitometry relies on rigid lane division and geometric assumptions, making it fragile against common artifacts like lane distortion, smiling effects, and complex backgrounds. While deep learning offers automation potential, existing solutions often lack precise quantification capabilities or require extensive preprocessing. To address these limitations, we propose an end-to-end, fully automated band segmentation framework based on the lightweight YOLO11n architecture.

Trained on 150 intrinsic fluorescence imaging images and fine-tuned on a small heterogeneous dataset (n = 50), the model utilizes a transfer learning strategy to ensure compatibility across Coomassie Brilliant Blue, silver, and fluorescence staining. By processing high-resolution inputs (1280 px) without preprocessing, the YOLO11-Seg model achieves a segmentation mAP50 of 0.947 with a rapid latency time of 3.1 ms, significantly outperforming U-Net and profile-based methods in resolving faint and distorted bands. Crucially, the method employs pixel-level masks rather than bounding boxes for quantification, effectively excluding background noise. This method demonstrates excellent linearity between integrated intensity and protein concentration (R² = 0.9964) with superior variability (CV as low as 3.3%). Integrated into the custom "Seg Lab" software, the workflow reduces per-gel analysis and visualization time from 4 min (manual workflow) to ∼1 s.

This study presents a robust, objective, and high-throughput alternative to traditional densitometry. By eliminating the dependence on lane division and manual intervention, the framework resolves the long-standing trade-off between analysis speed and quantification precision. Its lightweight design and proven cross-stain generalizability make it a practical, accessible tool for routine large-scale proteomics and analytical chemistry workflows.