AFJOG

African Journal of Obstetrics and Gynaecology | Volume 2 | Issue 2 | 2024 | 50 African Journal of Obstetrics and Gynaecology | Volume 2 | Issue 2 | 2024 | Abstracts AFJOG ABSTRACTS ABSTRACTTITLE: 193: Variables associated with HIV Viraemia at the time of delivery In The Department of Obstetrics and Gynaecology in Kalafong Provincial Tertiary Hospital AUTHORS: Dr Gabriel Kesekile gkesekile@gmail.com Gabriel Kesekile 1 , Felicia Molokoane 2 , A Mazanderani 3 1 University of Pretoria 2 University of Pretoria, Kalafong and Steve Biko 3 Department of Virology Introduction: Pregnancy-related infections are a significant cause of maternal and neonatal morbidity and mortality in South Africa. HIV infection is one of the driving forces behind these infections, more so in patients who have high viral loads. Understanding variables associated with high viral loads in pregnant women is of paramount importance in the fight against this pandemic and, ultimately, reduction in maternal and neonatal mortality and morbidity. Methods: The study's primary objectives were to describe maternal characteristics associated with the non-suppressed viral load during delivery. In addition, the neonatal outcomes of virally suppressed mothers were compared to those of virally non-suppressed mothers. A retrospective analytical observational study used historical patient data at the Kalafong Provincial Tertiary Hospital (KPTH) maternity unit. Records of women living with HIV (WLHIV) who delivered in KPTH between April 2018 and March 2019 were analyzed. Results: Out of 210 patients analyzed, 60% (n=126) had suppressed viral loads. WLHIV aged 35 years and older were more likely to achieve viral suppression (p = 0.002). Among the patients, 98% (n=206) remained on the first line of ART, with 60% (n=125) of these patients having suppressed viral loads. Of those on ART for over three months (87%, n=183), 67% (n=124) had suppressed viral loads. South Africans showed higher viral suppression rates (p = 0.001). Notably, patients who missed antenatal follow-up visits had non-suppressed viral loads. Neonates born to mothers with high viral loads had increased neonatal intensive care unit (NICU) admissions and stayed longer in the NICU compared to those with suppressed viral loads. Conclusion: The study highlights critical determinants of non-suppressed viral load among pregnant WLHIV. WLHIV on ART for less than three months, aged below 35 years, and with infrequent antenatal follow-ups were identified as higher-risk groups for non-suppressed viral load at delivery. ABSTRACTTITLE: 208: Predicting sperm DNA fragmentation: A non- destructive, stain-free approach using machine learning AUTHORS: Mr Ifthakaar Shaik ifthakaar.shaik@vitruvianmd.com Ifthakaar Shaik 1 1 VitruvianMD Introduction: The analysis of human sperm, traditionally a manual process, requires a specialised skill set that spans from wet slide preparation to advanced decision-making, including estimations of a patient’s fertility potential. Recent studies underscore the value of DNA fragmentation as a reliable indicator of male fertility, with Ghasemzadeh et al. [1] demonstrating its correlation with traditional fertility measures. Furthermore, Tang et al. [2] have identified DNA fragmentation as predictive of pregnancy loss, fertilization failure, and adverse outcomes in assisted reproduction technologies (ART). However, the chemical analysis of DNA fragmentation not only demands expensive equipment and chemicals but also involves time-consuming procedures and highly specialized skills. The ability to conduct digital sperm analysis using machine learning by accurately predicting DNA fragmentation from unstained brightfield images substantially reduces cost base, skill requirements and time. Methods: Our research evaluates the integrity of sperm DNA using three principal assays: Chromomycin A3 (CMA-3), the Acridine Orange test (AO), and the Sperm Chromatin Dispersion (SCD) test. CMA-3 detects protamine deficiency, indicating inadequate DNA packaging in spermatozoa. The AO test quantifies DNA strand breaks, serving as a marker for DNA fragmentation and potential apoptosis. The Halo sperm test (SCD) assesses DNA fragmentation, providing insights into sperm DNA structural integrity. Collectively, these assays offer comprehensive insights into sperm DNA condition, crucial for understanding male fertility. Studies by Hamidi et al. and Ghasemzadeh et al. indicate the necessity of multiple tests for a thorough analysis of sperm integrity. Additionally, the SCD test evaluates sperm DNA susceptibility to denaturation, directly assessing DNA stability and integrity. Fernandez et al. demonstrated the SCD test's utility in detecting subclinical DNA damage, often missed by other assays, highlighting its importance in sperm quality evaluation in clinical settings. Manual sperm assessment faces challenges in subjectivity, repeatability, and auditability. To address these, machine learning techniques, including convolutional deep neural networks and genetic neural architectures, have been employed for automated sperm analysis, focusing on sperm morphology. The field of AI is increasingly prevalent in fertility research, with reviews documenting recent advancements in AI applications within this domain. Results: We have developed and trained both deep convolutional neural networks and ensemble networks to accurately predict the DNA fragmentation index as indicated by CMA-3 and the AO assay. Our results demonstrate that these neural networks correlate strongly with the observed fluorescence intensity in sperm DNA fragmentation assessments using CMA-3 or AO. The SCD test, which measures DNA fragmentation through a different methodology involving physical dispersion of DNA loops, does not lend itself to per-sperm comparative analysis before and after the test. Consequently, our AI models showed poor correlation with SCD test results, highlighting the challenges in automating assessments across differing test methodologies. Conclusion: The rising costs, skill shortage and limited access to infertility diagnosis and treatment, presents a clear case for ML intervention. The ability to reproduce multiple