Optical spectroscopic devices have historically been too expensive or not portable enough to take full advantage of their abilities to offer real-time, on-site, objective results, especially in the developing world. Recent advancements toward smaller and cheaper hardware, especially in the visible and near infrared (NIR) ranges, could enable widespread use in low resource settings, down to a rural health clinic or at the individual farm level. We recently designed and tested a spectroscopic device with these goals in mind. It is based on an initial commercial version of a low cost MEMS spectral detection chip operating in the NIR, or more properly short wave infrared (SWIR) region. Custom optics, electronics, and mechanical designs were created to produce a complete handheld system capable of operation in the lab or in the field. Initial lab testing indicated excellent reproducibility both within and between five different devices. We have verified desired performance (e.g. acceptable signal to noise for target integration times, spectral features equivalent to lab-grade devices, etc.) for applications including pharmaceutical analysis and for analyzing multiple agricultural materials, including soils, plants, fertilizers, and manures. We have also developed a custom mobile app to accompany the devices in upcoming field testing, which will validate their performance in realistic settings in sub-Saharan Africa.
Adulteration of milk for economic gains is a widespread issue throughout the developing world that can have far-reaching health and nutritional impacts. Milk analysis technologies, such as infrared spectroscopy, can screen for adulteration, but the cost of these technologies has prohibited their use in low resource settings. Recent developments in infrared and Raman spectroscopy hardware have led to commercially available low-cost devices. In this work, we evaluated the performance of two such spectrometers in detecting and quantifying the presence of milk adulterants. Five common adulterants – ammonium sulfate, melamine, sodium bicarbonate, sucrose, and urea, were spiked into five different raw cow and goat milk samples at different concentrations. Collected MIR and Raman spectra were analyzed using partial least squares regression. The limit of detection (LOD) for each adulterant was determined to be in the range of 0.04 to 0.28% (400 to 2800 ppm) using MIR spectroscopy. Raman spectroscopy showed similar LOD’s for some of the adulterants, notably those with strong amine group signals, and slightly higher LOD’s (up to 1.0%) for other molecules. Overall, the LODs were comparable to other spectroscopic milk analyzers on the market, and they were within the economically relevant concentration range of 100 to 4000 ppm. These lower cost spectroscopic devices therefore appear to hold promise for use in low resource settings.
Cervical cancer is the fourth most common cancer among women worldwide and is especially prevalent in low resource settings due to lack of screening and treatment options. Visual inspection with acetic acid (VIA) is a widespread and cost-effective screening method for cervical pre-cancer lesions, but accuracy depends on the experience level of the health worker. Digital cervicography, capturing images of the cervix, enables review by an off-site expert or potentially a machine learning algorithm. These reviews require images of sufficient quality. However, image quality varies greatly across users. A novel algorithm was developed to evaluate the sharpness of images captured with the MobileODT’s digital cervicography device (EVA System), in order to, eventually provide feedback to the health worker. The key challenges are that the algorithm evaluates only a single image of each cervix, it needs to be robust to the variability in cervix images and fast enough to run in real time on a mobile device, and the machine learning model needs to be small enough to fit on a mobile device’s memory, train on a small imbalanced dataset and run in real-time. In this paper, the focus scores of a preprocessed image and a Gaussian-blurred version of the image are calculated using established methods and used as features. A feature selection metric is proposed to select the top features which were then used in a random forest classifier to produce the final focus score. The resulting model, based on nine calculated focus scores, achieved significantly better accuracy than any single focus measure when tested on a holdout set of images. The area under the receiver operating characteristics curve was 0.9459.