Measuring the performance of computer vision artificial intelligence to interpret images of HIV self-testing results.

Introduction: HIV self-testing (HIVST) is highly sensitive and specific, addresses known barriers to HIV testing (such as stigma), and is recommended by the World Health Organization as a testing option for the delivery of HIV pre-exposure prophylaxis (PrEP). Nevertheless, HIVST remains underutilized as a diagnostic tool in community-based, differentiated HIV service delivery models, possibly due to concerns about result misinterpretation, which could lead to inadvertent onward transmission of HIV, delays in antiretroviral therapy (ART) initiation, and incorrect initiation on PrEP. Ensuring that HIVST results are accurately interpreted for correct clinical decisions will be critical to maximizing HIVST's potential. Early evidence from a few small pilot studies suggests that artificial intelligence (AI) computer vision and machine learning could potentially assist with this task. As part of a broader study that task-shifted HIV testing to a new setting and cadre of healthcare provider (pharmaceutical technologists at private pharmacies) in Kenya, we sought to understand how well AI technology performed at interpreting HIVST results. Methods: At 20 private pharmacies in Kisumu, Kenya, we offered free blood-based HIVST to clients ≥18 years purchasing products indicative of sexual activity (e.g., condoms). Trained pharmacy providers assisted clients with HIVST (as needed), photographed the completed HIVST, and uploaded the photo to a web-based platform. In real time, each self-test was interpreted independently by the (1) client and (2) pharmacy provider, with the HIVST images subsequently interpreted by (3) an AI algorithm (trained on lab-captured images of HIVST results) and (4) an expert panel of three HIVST readers. Using the expert panel's determination as the ground truth, we calculated the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for HIVST result interpretation for the AI algorithm as well as for pharmacy clients and providers, for comparison. Results: From March to June 2022, we screened 1,691 pharmacy clients and enrolled 1,500 in the study. All clients completed HIVST. Among 854 clients whose HIVST images were of sufficient quality to be interpretable by the AI algorithm, 63% (540/854) were female, median age was 26 years (interquartile range: 22-31), and 39% (335/855) reported casual sexual partners. The expert panel identified 94.9% (808/854) of HIVST images as HIV-negative, 5.1% (44/854) as HIV-positive, and 0.2% (2/854) as indeterminant. The AI algorithm demonstrated perfect sensitivity (100%), perfect NPV (100%), and 98.8% specificity, and 81.5% PPV (81.5%) due to seven false-positive results. By comparison, pharmacy clients and providers demonstrated lower sensitivity (93.2% and 97.7% respectively) and NPV (99.6% and 99.9% respectively) but perfect specificity (100%) and perfect PPV (100%). Conclusions: AI computer vision technology shows promise as a tool for providing additional quality assurance of HIV testing, particularly for catching Type II error (false-negative test interpretations) committed by human end-users. We discuss possible use cases for this technology to support differentiated HIV service delivery and identify areas for future research that is needed to assess the potential impacts-both positive and negative-of deploying this technology in real-world HIV service delivery settings.

Online HIV prophylaxis delivery: Protocol for the ePrEP Kenya pilot study.

Background: Online pharmacies in Kenya provide sexual and reproductive health products (e.g., HIV self-testing, contraception) and could be leveraged to increase the reach of HIV pre-exposure and post-exposure prophylaxis (PrEP/PEP) to populations who do not frequently attend health facilities. To date, evidence is limited for operationalizing online PrEP/PEP delivery and the type of populations reached with this differential service delivery model. Methods: The ePrEP Kenya Pilot will deliver daily oral PrEP and PEP via MYDAWA, a private online pharmacy retailer, to clients in Nairobi for 18 months. Potential clients will obtain information about PrEP/PEP on MYDAWA's sexual wellness page and self-screen for HIV risk. Individuals ≥18 years, identified as at HIV risk, and willing to pay for a blood-based HIV self-test and PrEP/PEP delivery will be eligible for enrollment. To continue with online PrEP/PEP initiation, eligible clients will purchase a blood-based HIV self-test for 250 KES (~USD 2) [delivered to their setting of choice for 99 KES (~USD 1)], upload an image of their self-test result, and attend a telemedicine visit with a MYDAWA provider. During the telemedicine visit, providers will screen clients for PrEP/PEP eligibility, including clinical concerns (e.g., kidney disease), discuss self-test results, and complete counseling on PrEP/PEP use and safety. Providers will refer clients who self-test HIV positive or report any existing medical conditions to the appropriate services at healthcare facilities that meet their preferences. Eligible clients will be prescribed PrEP (30-day PrEP supply at initiation; 90-day PrEP supply at follow-up visits) or PEP (28-day supply) for free and have it delivered for 99 KES (~USD 1). We will measure PrEP and PEP initiation among eligible clients, PEP-to-PrEP transition, PrEP continuation, and implementation outcomes (e.g., feasibility, acceptability, and costs). Discussion: Establishing pathways to increase PrEP and PEP access is crucial to help curb new HIV infections in settings with high HIV prevalence. The findings from this study will provide evidence on the implementation of online pharmacy PrEP and PEP service delivery that can help inform guidelines in Kenya and similar settings.

Differential oblique angle spectroscopy of the oral epithelium.

Increasing evidence suggests that inflammation may contribute to the process of carcinogenesis. This is the basis of several clinical trials evaluating potential chemopreventive drugs. These trials require quantitative assessments of inflammation, which, for the oral epithelium, are traditionally provided by histopathological evaluation. To reduce patient discomfort and morbidity of tissue biopsy procedures, we develop a noninvasive alternative using diffuse reflectance spectroscopy to measure epithelial thickness as an index of tissue inflammation. Although any optical system has the potential for probing near-surface structures, traditional methods of accounting for scattering of photons are generally invalid for typical epithelial thicknesses. We develop a single-scattering theory that is valid for typical epithelial thicknesses. The theory accurately predicts a distinctive feature that can be used to quantify epithelial thickness given intensity measurements with sources at two different angles relative to the tissue surface. This differential measure approach has acute sensitivity to small, layer-related changes in scattering coefficients. To assess the capability of our method to quantify epithelial thickness, detailed Monte Carlo simulations and measurements on phantom models of a two-layered structure are performed. The results show that the intensity ratio maximum feature can be used to quantify epithelial thickness with an error less than 30% despite fourfold changes in scattering coefficients and 10-fold changes in absorption coefficients. An initial study using a simple two-source, four-detector probe on patients shows that the technique has promise. We believe that this new method will perform well on patients with diverse tissue optical characteristics and therefore be of practical clinical value for quantifying epithelial thickness in vivo.

Noninvasive infrared imaging for quantitative assessment of tumor vasculature and response to therapy.

In this study we are investigating three infrared imaging techniques, thermography, multispectral imaging and Laser Doppler imaging (LDI) to assess parameters of vascularity in lesions of Kaposi's sarcoma (KS) and response to therapy. Thermography, multispectral imaging and LDI were recorded over the lesion and compare to normal skin either adjacent to the lesion or on the contralateral side. The KS lesions generally had increased temperature, blood volume (as measured by multispectral imaging) and blood flux (as measured by LDI) as compare to normal skin. After the treatment with experimental antiKS drug, temperature, blood volume and blood flow of the lesion were significantly reduced from the baseline. These techniques hold promise to assess physiological parameter in KS lesion and their changes with therapy.

Tissue characterization by quantitative optical imaging methods.

Optical methods have a long history in the field of medical diagnosis. The biomolecular specificity possible with optical methods has been particularly valuable in microscopy and histopathology while in vivo imaging of deep structures has traditionally been the domain of X-ray and MRI. The use of optical methods in deep tissue has been limited by multiple-scattering which blurs or distorts the optical signal. New stochastic methods which account for multiple scattering have been developed that are extending the usefulness of optical methods deep into tissue. In optical mammography, photons may travel through 10 cm of tissue before arriving at the detector. We have developed a method for quantifying parameters of anomalous sites in breast tissue that may be used for functional characterization of tumors. In other work presented here, we are developing fluorescence based methods to detect and monitor tumor status. The immune response to a tumor is a target for fluorescently labeled specific antibodies. We have developed a method to localize the tumor site using CW fluorescence. Additionally, we have developed a method which uses time-resolved data and capitalizes on probe lifetime sensitivity to metabolic parameters such as pH and temperature to obtain functional information from the tumor site.

Analytical calculation of the mean time spent by photons inside an absorptive inclusion embedded in a highly scattering medium.

The mean time spent by photons inside a nonlocalized optically abnormal embedded inclusion has been derived analytically. The accuracy of the results has been tested against Monte Carlo and experimental data. We show that for quantification of the absorption coefficient of absorptive inclusions, a corrective factor that takes into account the size of the inclusion is needed. This finding suggests that perturbation methods derived for very small inclusions which are used in inverse algorithms require a corrective factor to adequately quantify the differential absorption coefficient of nonlocalized targets embedded in optically turbid media.

Quantification of optical properties of a breast tumor using random walk theory.

For the first time we use a random walk methodology based on time-dependent contrast functions to quantify the optical properties of breast tumors (invasive ductal carcinoma) of two patients. Previously this theoretical approach was successfully applied for analysis of embedded objects in several phantoms. Data analysis was performed on distributions of times of flight for photons transmitted through the breast which were recorded in vivo using a time-domain scanning mammograph at 670 and 785 nm. The size of the tumors, their optical properties, and those of the surrounding tissue were reconstructed at both wavelengths. The tumors showed increased absorption and scattering. From the absorption coefficients at both wavelengths blood oxygen saturation was estimated for the tumors and the surrounding tissue.