Technologies
This technology is a machine learning platform that automatically interprets rapid diagnostic tests from images and uses self-supervised pre-training to adapt to new test kits with minimal data.
Current point-of-care and at-home rapid diagnostic tests rely on users to visually interpret faint or ambiguous test lines, leading to high error rates, misguided clinical decisions, and inconsistent results. Existing digital interpretation tools typically require large, kit-specific training datasets and fail to generalize to new or updated test formats. As rapid diagnostic kits continue to evolve, there are currently no scalable methods to ensure accurate and standardized interpretation across diverse kit designs.
This technology is an automated image interpretation system for rapid diagnostic tests that uses a neural network pre-trained to recognize key visual features of test kits. In its pre-training stage, the model learns general structural patterns by converting test kit images to grayscale, extracting features such as edges, and reconstructing these simplified images through a feature-extraction and decoding process. By minimizing reconstructions during training, the system refines its feature extractor to capture line-based patterns and other consistent visual characteristics across different test kit designs. This structure enables the model to adapt efficiently to new rapid test kits with limited data. This technology can be implemented on a smartphone, providing accurate interpretation in a portable, user-accessible format.
This technology has been validated using a large dataset that spans multiple rapid test kits.
Patent Pending (US20230274538)
IR CU21034
Licensing Contact: Dovina Qu