Article • Point-of-care diagnostics

Improving women’s health in remote regions with digital pathology

Point-of-care diagnostics based on a combination of mobile-sized scanners and artificial intelligence (AI) are helping save the lives of women in low-resource settings. The AI technique is being applied in Kenya and Tanzania to deliver screening for cervical cancer – now the leading cause of cancer-related deaths in women in that region and a bigger cause of death than childbirth.

Report: Mark Nicholls

Behind the digital pathology initiative is a team from Karolinska Institutet and Uppsala University in Sweden; the Institute for Molecular Medicine (FIMM) at the University of Helsinki in Finland; Kinondo Kwetu Hospital in Kenya; and Muhimbili University of Health and Allied Sciences in Tanzania.

The project, which sees microscope scanners wirelessly connected via mobile networks for deep learning-based image analysis, was outlined at the Digital Pathology and AI congress in London in December by Professor Nina Linder. She explained that in parts of Africa there are less than four pathologists for every million people, underlining the importance of the drive towards automated diagnostics. Linder is a Principal Investigator in the studies, along with Professor Johan Lundin from Karolinska Institutet and FIMM and Professor Andreas Mårtensson from Uppsala University.

The solution from the Scandinavian team uses minimal infrastructure for point-of-care diagnostics and sees a high-quality sample collected from a local medical centre with the Pap smear digitized with a low-cost scanner and transferred to a cloud server over mobile network for remote diagnosis by a pathologist, AI, or a combination of both. The result is then sent back to southern Kenya or Tanzania. Linder pointed out several advantages of point-of-care AI-based diagnostics, such as allowing remoted consultations, decreasing workforce burden, monitoring disease outbreaks and storage of image data.

First medical AI training centre in Africa

The team hopes the project will help tackle cervical cancer, which affects relatively young women and is preventable with early detection. Currently, slow deployment of vaccines to protect against the human papilloma virus (HPV) in sub-Saharan Africa laed experts to predict a doubling in mortality by 2030.

Digital pathology can be applied in resource-limited settings using a minimal infrastructure approach at the primary health care level with AI-based diagnostic assistants likely to improve access to high-quality diagnostics in these areas
Nina Linder

The study project, which includes a validation and verification element to compare the accuracy of the AI method to diagnostics by a pathologist, saw the first cohort of 740 HIV-positive women scanned in 2019, followed by a second cervical cancer screening study in 2022/23. In 2024, a third study focused on 2,500 women from the general population.

The work is primarily centred around Kinondo Kwetu Hospital in Kwale county, which sees around 130,000 patients a year. At the hospital, a core study team of two nurses performs participant counselling and takes Pap smear samples; sample staining, quality control and scanning is handled by laboratory technicians. The onsite team is completed by a data manager; consultant pathologist and gynaecologist, in what is the first medical AI training centre in Africa.

To train the algorithm, Linder explained that each digitized slide corresponds to a microscope slide with annotations by a researcher and cytotechnologist. ‘More than 16,000 regions of interest on 350 slides were manually annotated and these included areas of normal cervical morphology and various degrees of atypia. Training the algorithm used 30,000 iterations.’

The results for screening cervical abnormalities with AI to date have been promising: the sensitivity for detection of atypia was high (96-100%); higher specificity for high-grade lesions (93-99%) than for low-grade lesions (82-86%); and the deep learning system found 93/361 slides with atypia (29 high grade and 64 low grade), corresponding to about two times more than the cytotechnician. ‘None of the patient samples manually classified as high-grade were incorrectly classified as negative,’ added Linder, ‘and all women with abnormal Pap smears received treatment in line with local guidelines.’

Equitable access

Central to the 2024 study was a free health camp as well as a clinical study for more than 1600 women, with some travelling for two days to attend. The project, which closely follows WHO guidance on the use of AI for health, has received media attention in Kenya.

Additionally, Linder heads the topic group on AI for point-of-care diagnostics within the United Nations Global Initiative on AI for Health. ‘Key messages from our clinical studies are that it is feasible to use AI in a real-world setting for screening of cervical cancer and it has enormous potential for saving women’s lives,’ she said. ‘Digital pathology can be applied in resource-limited settings using a minimal infrastructure approach at the primary health care level with AI-based diagnostic assistants likely to improve access to high-quality diagnostics in these areas. The diagnostic system provides a tool for access to advanced diagnostics at the point-of-care and is a significant step towards more equitable and sustainable access to high quality diagnostics in a global setting.’

Profile:

Nina Linder, MD, PhD, is a Guest Professor of Medical Diagnostic Artificial Intelligence at the Department of Women’s and Children’s Health, Uppsala University, Sweden. She is also a researcher at the Institute for Molecular Medicine (FIMM) at the University of Helsinki in Finland. Her research focus is on the development of novel AI-based solutions for cancer and infectious disease diagnostics.

05.02.2025