Deep Learning on chromatographic data for Segmentation and Sensitive Analysis.

Area of Science:

• Biomedical Engineering
• Medical Diagnostics
• Image Analysis

Background:

• Lateral flow immunoassay (LFIA) is crucial for point-of-care testing.
• Current image analysis tools struggle with LFIA image quality due to process pollution and irregular backgrounds, impacting detection accuracy.
• Machine learning offers advanced solutions for image recognition and processing in biochemical analysis.

Purpose of the Study:

• To develop and evaluate a U-Net based convolutional neural network (CNN) for quantitative analysis of LFIA images.
• To accurately segment and extract target signals (T-/C-lines) from LFIA images, enabling precise quantification.
• To improve the accuracy and reliability of LFIA detection, especially for weak positive signals and multi-target scenarios.

Main Methods:

• Image preprocessing techniques including graying, binarization, and labeling were applied to LFIA images.
• A U-Net convolutional neural network (CNN) architecture was employed for image segmentation and signal extraction.
• Performance was evaluated using metrics such as peak signal-to-noise ratio and intersection-over-union (IoU).

Main Results:

• The U-Net model achieved a peak signal-to-noise ratio of 22.4 dB, a significant improvement over existing methods (4 dB).
• The intersection-over-union (IoU) metric exceeded 93%, demonstrating high accuracy in segmenting target areas.
• The method effectively processed single- and multi-target LFIA images, showing potential for weak positive signal detection.

Conclusions:

• The proposed U-Net deep learning approach provides a robust and effective method for quantitative analysis of LFIA images.
• This technique significantly enhances the accuracy of signal extraction and detection in LFIA, addressing limitations of current tools.
• The deep learning method holds substantial potential for advancing rapid detection devices, systems, and biological image analysis in diagnostics.