Using infrared spectroscopy and machine-learning to improve malaria diagnosis in low-income communities

Objectives

This project will validate the potential of mid-infrared spectroscopy & deep learning (MIR-ML) for malaria screening in areas with varying prevalence rates, and demonstrate the integration of this approach to improve case detection and care. This grant will support all the activities and will cover some personnel costs.

Specific objectives:

1. Evaluate sensitivity and specificity of the MIR-ML approach in areas with different levels of transmission. Since Tanzania has ~1 health facility/village, the spatial variations can allow to assess the validity of the new approach in areas of varying parasite densities and transmission intensities.
2. Assess the diagnostic thresholds of the MIR-ML approach, and its suitability in areas of varying endemicity levels: First, they will aliquot malaria-positive specimen to achieve parasite density ranges of <10/μL up to >10,000/μL, then examine accuracies of MIR-ML at those ranges. Second, they will evaluate statistical correlations between the sensitivity of the ML-MIR approach and natural infection intensities observed in the field, so as to detect thresholds below or beyond which such correlations decay. By extension, this study could also enable in-depth evaluation of the biological basis of observed test results. It can extracted the most dominant peaks on the mid-infrared spectra, which will form the basis for studying the biological basis of “infected” and “non-infected” signals.
3. Develop an online cloud-based system that integrates data collected from different malaria diagnostic stations in candidate villages, and makes this data available for iterative training of the deep learning algorithms. The diagnostic accuracies by any of the contributing sites will be constantly increased. Currently, clinicians working in nearby health facilities do not have a way to communicate diagnostic outcomes among themselves, yet this is an important aspect of the surveillance-response initiative. This approach will therefore help other facilities regardless of diagnostic methods used, to be aware of new infections in their areas of jurisdiction or neighboring areas. This will improve human-machine interactions and overall diagnostic experiences.