Authors

M Lemin²; C Roberts²; A Bustinduy¹;  ¹ London School of Hygiene and Tropical Medicine, UK;  ² Clinical Research Department, London School of Hygiene and Tropical Medicine, UK

Discussion

The exponential growth in artificial intelligence-powered computer vision has led to significant developments in how we diagnose disease. There are already more than fifteen machine learning algorithms that have been developed for the identification of cervical cancer in colposcope images, although as of mid-2024 none have been reported to be regularly used in a clinical setting. Work is underway to develop a computer vision software platform for the enhanced detection of female genital schistosomiasis (FGS) in cervical images, using over 20,000 images from multiple sources. A basic binary classification (FGS positive vs. FGS negative) convolutional neural network (CNN) has already been written and tested, yielding a high specificity (89%) but low sensitivity (52%). These results provide an encouraging baseline for the ongoing development of more complex algorithms, which will include Bayesian modelling to ensure the consideration of individual risk factors. These methods show significant promise for application in low-resource settings, however, work is needed to ensure their successful implementation.