Artificial intelligence for screening of cervical cancer and infectious diseases at the primary healthcare level in resource-limited settings – the MoMic Study – Johan Lundin

Introduction: Access to accurate and timely medical diagnostics remains a major global health challenge, particularly in low-resource settings where shortages of trained professionals and laboratory infrastructure hinder effective healthcare delivery. Advances in artificial intelligence (AI) offer the potential to bridge this gap by enabling automated and cost-effective diagnostic solutions. We have conducted a series of clinical studies at the primary healthcare level in resource-limited settings in East Africa to assess the diagnostic accuracy of AI-based methods for cervical cancer screening and diagnostics of soil-transmitted helminth infections.

Methods: Whole-slide microscopy samples are digitized with portable microscope scanners that are wirelessly connected via mobile networks for AI-based image analysis in a cloud environment using deep learning algorithms. The method represent a minimal infrastructure approach that can be used in a primary healthcare laboratory and includes a verification step, where a human expert verifies AI-based findings to reach a sample level diagnosis.

Results: We have screened 3,802 women for precancerous lesions in Pap smears, and 2,258 school children for soil-transmitted helminth eggs in stool samples at five study sites in Kenya and Tanzania. The method showed a sensitivity for detection of atypical cervical cells of 96-100%, with higher specificity for high grade lesions (93-99%) than for low grade lesions (82-86%). In a validation on prospectively collected samples at the same hospital the sensitivity for high grade atypia was 100% with a specificity of 84%. Assessment of samples collected in 2024 is ongoing. The sensitivity of the expert verified AI for the soil-transmitted helminths A. lumbricoides, T. trichiura and hookworm was 100%, 94% and 92%, respectively.

Conclusions: AI-powered image analysis can assist in cancer and infectious disease screening and diagnostics, reducing dependency on scarce human expertise. However, significant challenges remain, including access to data, regulatory hurdles, and the need for validation in diverse populations. Additionally, ethical considerations around AI deployment, infrastructure requirements, and sustainability must be addressed to ensure equitable implementation. This presentation explores both the opportunities and limitations of AI-based diagnostics, emphasizing the importance of global collaboration, robust validation, and context-specific solutions.

About Johan Lundin
Johan Lundin, MD, PhD is a Professor of Medical Technology at Karolinska Institutet, Stockholm, Sweden and a Research Director and Group Leader at the Institute for Molecular Medicine Finland (FIMM), University of Helsinki, Finland. He is a medical doctor by training and his overall research aims are to study the use of digital technologies and artificial intelligence (AI) for improvement of diagnostics and care of the individual patient. In addition to the research, Dr. Lundin has together with his co-workers developed technologies for diagnostic decision support, for example cloud-based and mobile solutions that allow the diagnostic process to be performed using AI-supported analysis in both high- and low-resource settings.