Mobile-ELISA: A machine learning-enhanced smartphone platform for ELISA diagnostics in veterinary medicine.

Abstract

This dissertation presents the development and validation of MOBILE-ELISA that uses computer vision and machine learning to improve the accessibility and accurate interpretation of ELISA diagnostics in veterinary medicine. Traditional ELISA methods, though highly sensitive and specific, are limited by their reliance on expensive, bulky spectrophotometers and time consuming procedures, making them impractical for resource-limited settings. To address these challenges, this research implements smartphone technology together with advanced machine learning algorithms to develop a portable, cost-effective alternative. The study used a comparative experimental design, analysing 3,168 wells from 33 ELISA plates using both smartphone imaging and conventional spectrophotometry methods. Two feature extraction pipeline i.e. OpenCV (Hough Circles) and the RFDETR deep learning model were evaluated for well detection. RFDETR outperformed OpenCV because it achieved 93% accuracy and maintained robustness when dealing with different imaging conditions. For optical density (OD) prediction from extracted RGB values of each ELISA well, the stacked ensemble machine learning model achieved moderate correlation with spectrophotometer readings through its R² value of 0.6931 and MSE value of 0.1535, and the diagnostic agreement analysis yielded an AUC of 0.82, with 78% sensitivity and 75% specificity when using the clinical cut off of OD ≥0.4. Key findings show the platform's potential for decentralized screening applications settings because it reduces costs by 30 times compared to spectrophotometer use. Future directions include expanding the training dataset, standardized imaging protocols, and pathogen specific validation across veterinary medicine. By integrating smartphone technology with machine learning, MOBILE ELISA bridges critical gaps in veterinary diagnostics, providing rapid, on-site testing and supporting global efforts toward equitable animal health monitoring