Artificial intelligence-driven mobile interpretation of a semi-quantitative cryptococcal antigen lateral flow assay

Objectives: Cryptococcosis remains a severe global health concern, underscoring the urgent need for rapid and reliable diagnostic solutions. Point-of-care tests (POCTs), such as the cryptococcal antigen semi-quantitative (CrAgSQ) lateral flow assay (LFA), offer promise in addressing this challenge. However, their subjective interpretation poses a limitation. Our objectives encompass the development and validation of a digital platform based on Artificial Intelligence (AI), assessing its semi-quantitative LFA interpretation performance, and exploring its potential to quantify CrAg concentrations directly from LFA images.

Methods: We tested 53 cryptococcal antigen (CrAg) concentrations spanning from 0 to 5000 ng/ml. A total of 318 CrAgSQ LFAs were inoculated and systematically photographed twice, employing two distinct smartphones, resulting in a dataset of 1272 images. We developed an AI algorithm designed for the automated interpretation of CrAgSQ LFAs. Concurrently, we explored the relationship between quantified test line intensities and CrAg concentrations.

Results: Our algorithm surpasses visual reading in sensitivity, and shows fewer discrepancies (p < 0.0001). The system exhibited capability of predicting CrAg concentrations exclusively based on a photograph of the LFA (Pearson correlation coefficient of 0.85).

Conclusions: This technology's adaptability for various LFAs suggests broader applications. AI-driven interpretations have transformative potential, revolutionizing cryptococcosis diagnosis, offering standardized, reliable, and efficient POCT results.