Smartphone-Compatible Colorimetric Detection of CA19-9 Using Melanin Nanoparticles and Deep Learning

Area of Science:

• Biomaterials Science
• Nanotechnology
• Machine Learning

Background:

• Paper-based colorimetric biosensors offer low-cost diagnostics but face environmental and interpretation challenges.
• Conventional metal nanomaterials raise ecological concerns, and visual interpretation can be subjective.

Purpose of the Study:

• To develop an environmentally sustainable and analytically robust colorimetric quantification platform.
• To integrate naturally derived melanin nanoparticles (MNPs) with machine learning for biomarker detection.

Main Methods:

• Utilized melanin nanoparticles (MNPs) for a paper-based biosensor detecting CA19-9 biomarker.
• Employed machine learning for automated image analysis, including U-Net for region segmentation and XGBoost for regression.
• Quantified concentration-dependent color transitions from yellow to brown induced by MNPs upon target binding.

Main Results:

• The U-Net model achieved high segmentation accuracy (IoU: 0.9025 ± 0.0392).
• XGBoost regression model demonstrated strong predictive accuracy (MAPE: 17%) at low analyte concentrations.
• The platform provided accurate, reproducible, and objective quantification of colorimetric signals.

Conclusions:

• This approach offers a sustainable and scalable alternative for point-of-care diagnostics.
• Melanin nanoparticles combined with machine learning overcome limitations of traditional biosensors.
• The developed platform enhances objectivity and reliability in colorimetric quantification.