Chest X-rays are a vital diagnostic tool in the workup of many patients. Similar to most medical imaging modalities, they are profoundly multi-modal and are capable of visualising a variety of combinations of conditions. There is an ever pressing need for greater quantities of labelled images to drive forward the development of diagnostic tools; however, this is in direct opposition to concerns regarding patient confidentiality which constrains access through permission requests and ethics approvals. Previous work has sought to address these concerns by creating class-specific generative adversarial networks (GANs) that synthesise images to augment training data. These approaches cannot be scaled as they introduce computational trade offs between model size and class number which places fixed limits on the quality that such generates can achieve. We address this concern by introducing latent class optimisation which enables efficient, multi-modal sampling from a GAN and with which we synthesise a large archive of labelled generates. We apply a Progressive Growing GAN (PGGAN) to the task of unsupervised X-ray synthesis and have radiologists evaluate the clinical realism of the resultant samples. We provide an in depth review of the properties of varying pathologies seen on generates as well as an overview of the extent of disease diversity captured by the model. We validate the application of the Fréchet Inception Distance (FID) to measure the quality of X-ray generates and find that they are similar to other high-resolution tasks. We quantify X-ray clinical realism by asking radiologists to distinguish between real and fake scans and find that generates are more likely to be classed as real than by chance, but there is still progress required to achieve true realism. We confirm these findings by evaluating synthetic classification model performance on real scans. We conclude by discussing the limitations of PGGAN generates and how to achieve controllable, realistic generates going forward. We release our source code, model weights, and an archive of labelled generates.

Background: Infection with SARS-CoV-2 has shown to cause an increase in D-dimers, which correlate with severity and prognosis for in-hospital mortality. The B.1.617.2 (delta) variant is known to cause a raised D-dimer level, with data on D-dimers in the B.1.1.529 (omicron) variant being scarce.Objectives: To determine the effect of age, gender and SARS-CoV-2 variant on the D-dimer in South Africans admitted to tertiary medical centres from May 2021 to December 2021.Method: The study was performed retrospectively on 16 010 adult patients with a SARS-CoV-2 infection. Age, gender, SARS-CoV-2 PCR and D-dimer levels on admission were collected from two national laboratories. Admissions from 01 May 2021 to 31 October 2021 were classified as B.1.617.2, whereas admissions from 01 November 2021 to 23 December 2021 were classified as B.1.1.529 infections.Results: Omicron infections had a median D-dimer level of 0.54 µg/mL (95% CI: 0.32, 1.08, p  0.001). Multivariable regression analysis showed that infection with omicron had a 34.30% (95% CI: 28.97, 39.23) reduction in D-dimer values, compared with delta infections. Middle aged, aged and aged over 80 years had D-dimer results greater than the adult baseline (42.6%, 95% CI: 38.0, 47.3, 124.6%, 95% CI: 116.0, 133.7 and 216.1%, 95% CI: 199.5, 233.3). Males on average had a 7.1% (95% CI: 4.6, 9.6) lower D-dimer level than females.Conclusion: Infection with the B.1.1.529 variant, compared with B.1.617.2 variant, had significantly lower D-dimer levels, with age being a more significant predictor of D-dimer levels, than gender and SARS-CoV-2 variant of infection.Contribution: This study provides novel insight into the hypercoagulable impact of various SARS-CoV-2 variants, which can guide the management of patients.

Despite advances in reducing HIV-related mortality, persistently high HIV incidence rates are undermining global efforts to end the epidemic by 2030. The UNAIDS Fast-track targets as well as other preventative strategies, such as pre-exposure prophylaxis, have been identified as priority areas to reduce the ongoing transmission threatening to undermine recent progress. Accurate and granular risk prediction is critical for these campaigns but is often lacking in regions where the burden is highest. Owing to their ability to capture complex interactions between data, machine learning and artificial intelligence algorithms have proven effective at predicting the risk of HIV infection in both high resource and low resource settings. However, interpretability of these algorithms presents a challenge to the understanding and adoption of these algorithms. In this perspectives article, we provide an introduction to machine learning and discuss some of the important considerations when choosing the variables used in model development and when evaluating the performance of different machine learning algorithms, as well as the role emerging tools such as Shapely Additive Explanations may play in helping understand and decompose these models in the context of HIV. Finally, we discuss some of the potential public health and clinical use cases for such decomposed risk assessment models in directing testing and preventative interventions including pre-exposure prophylaxis, as well as highlight the potential integration synergies with algorithms that predict the risk of sexually transmitted infections and tuberculosis.

No abstract

BACKGROUND Automated machine guided tools could be a valuable complement to electronic health initiatives to screen for diseases as well as link patients to care. Using supervised learning where a machine is programmed with labelled data sets with a built-in desired output, such tools can be trained to predict how prone an individual is to being infected with a disease. This study explored the feasibility of using a machine guided tool to predict susceptibility to HIV infection. OBJECTIVE To evaluate the accuracy of a machine-guided risk assessment tool in assessing HIV risk in those believed to be negative or unaware of their HIV status. METHODS 593 participants were recruited from three different geographical locations. On enrolment, participants undertook their first visit where they answered a collection of questions on the machine-guided tool, before having two HIV rapid diagnostics tests performed on them. HIV negative participants were invited to a follow-up visit, where the process was repeated. The machine-guided tool evaluated the participants’ individual responses and generated risk-assessment scores or predictions which were used to assess the viability of machine-guided tool in the identification of HIV at-risk patients. RESULTS A great majority of participants were HIV negative and male (333/517, 64.4%). The median age of participants who were HIV negative was 26 years, whilst that of HIV positive participants was 32 years. KwaZulu-Natal yielded the largest number of patients (n=199). Only 36/517 (7.0%) HIV negative and 12/76 HIV (15.8%) positive patients had never been tested for HIV previously. The greatest predictors of HIV susceptibility were: age (P ≤ .001), Frequency of HIV testing (P ≤ .001), sex (P = .012), and vaginal insertive sexual activity (P = .002). Risk prediction models yielded AUROC values ranging from 77.78% to 81.72%. A boosted tree model with a cut-off value of 0.15 performed best with a sensitivity of 83% (95% CI 71 – 91), specificity of 71% (95% CI 67 – 76), and a negative predictive value of 97% (95% CI 94 – 98) in a hold-out dataset. CONCLUSIONS The machine guided tool risk prediction indicated a high degree of sero-positivity sensitivity. Therefore, it displays potential in identifying candidates at risk of contracting HIV or needing intervention. However, the extent of the machine-guided tool’s viability remains to be validated. CLINICALTRIAL South African National Clinical Trial Registry: DOH-27-042021-679