Artificial intelligence performance in testing microfluidics for point-of-care

Area of Science:

• Medical technology
• Artificial intelligence in diagnostics
• Microfluidics for healthcare

Background:

• Artificial intelligence (AI) is transforming medicine, particularly in diagnostics and patient care.
• AI excels in tasks like image segmentation and pattern recognition, integrating with existing healthcare platforms.
• While AI shows promise in microfluidics for point-of-care (POC) diagnostics, comparative studies of AI algorithms for microfluidic testing are lacking.

Purpose of the Study:

• To comparatively evaluate machine learning (ML) and deep learning (DL) AI models for bubble detection in microfluidic channels.
• To identify the best-performing AI algorithms for microfluidic testing under various imaging conditions.
• To assess the potential of AI, specifically DL models, for mobile POC diagnostic applications.

Main Methods:

• A model microfluidic system with a single channel containing 3D transparent objects (bubbles) was used.
• Six ML and nine DL models were tested across different background settings to classify the presence or absence of bubbles.
• Performance was evaluated using sensitivity, specificity, and Area Under the Curve (AUC).

Main Results:

• The random forest ML model achieved 95.52% sensitivity, 82.57% specificity, and 97% AUC, outperforming other ML algorithms.
• DenseNet169, a DL model suitable for mobile integration, demonstrated 92.63% sensitivity, 92.22% specificity, and 92% AUC.
• DenseNet169 integrated into a mobile POC system achieved high accuracy (>0.84) in challenging microfluidic testing conditions.

Conclusions:

• AI holds significant potential to revolutionize precision medicine through accurate and accessible diagnostics.
• The study highlights the effectiveness of specific AI models, like random forest and DenseNet169, for microfluidic bubble detection.
• Integrating AI into healthcare systems can enhance patient outcomes and streamline diagnostic processes, especially in POC settings.