Stakeholders’ perceptions of AI-assisted obstetric ultrasound in rural and urban Kenya: a qualitative study
This qualitative study in rural and urban Kenya reveals that while stakeholders view AI-assisted obstetric ultrasound with cautious optimism, its successful and ethical implementation requires addressing concerns about trust, data privacy, and equity through community co-design and alignment with existing health system realities.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Pregnancy is a time of profound hope and anxiety, where the simple question of whether a baby is growing well can determine the course of a life. In many parts of the world, the answer comes from an ultrasound scan, a safe and painless image that allows doctors to see inside the womb. However, in rural areas with few resources, these machines are often scarce, and the skilled technicians needed to read them are even harder to find. This gap leaves many mothers without critical information about their pregnancies. To bridge this divide, scientists are exploring the use of artificial intelligence, a field of computer science where machines learn to recognize patterns and make decisions, much like a human brain but powered by data. The idea is that a smart computer could help a less experienced health worker perform an ultrasound and interpret the results with high accuracy, bringing specialist-level care to remote villages. But before such technology can be rolled out, it must be accepted by the people it is meant to serve.
A team of researchers set out to understand how people in Kenya feel about this future possibility. They traveled to two very different places: the bustling capital city of Nairobi and the rural, coastal region of Kilifi. In these settings, they spoke with nearly ninety people, including pregnant women, their partners, doctors, nurses, community leaders, and local health officials. The researchers did not ask these people to evaluate a working machine, because the device had not been built yet. Instead, they used a simple drawing to explain the concept: a nurse using a small, portable scanner that would be guided by a computer to find problems in a pregnancy. The goal was to listen to the hopes, fears, and conditions these stakeholders would place on such a tool before it ever left the laboratory.
The conversations revealed a landscape of cautious optimism. Many participants, especially those in rural areas where ultrasound services are rare, saw the potential for this technology to be a lifeline. They imagined a world where a nurse in a remote dispensary could use a battery-powered device to check a baby's health without needing to send the mother on a long, expensive journey to a city hospital. They hoped it would help detect complications early, saving lives and reducing the number of mothers who rely on traditional birth attendants because they have no other option. For some, the idea of a machine that could help a minimally trained worker perform a complex task was a powerful step toward fairness, offering the same quality of care to a woman in a village as to one in a city.
Yet, this hope was tempered by deep and specific concerns. Trust was the central theme, and it was not given lightly. Many participants worried that the computer might make mistakes, especially if it had been trained on images from different people or places. They used a simple logic: if the machine learns from poor-quality pictures, it will give poor-quality answers. They insisted that for the tool to be trusted, it must be taught using images from local communities, so it understands the specific health realities of the people it serves. There was also a strong desire for human oversight. Participants did not want a machine to replace the doctor or the nurse; they wanted the computer to act as a helpful assistant, with a skilled human always there to double-check the results and explain them to the mother. The fear was that without a human to interpret the findings, a mother might receive a confusing or alarming message without anyone there to comfort her or guide her to the next step.
Ethical questions also surfaced, particularly regarding privacy and the use of data. The researchers explained that to build a smart system, they would need to collect and store thousands of ultrasound images to teach the computer. While many women were willing to share their images to help others, they had strict conditions. They demanded absolute anonymity, fearing that their private medical images could be stolen or misused. Some expressed discomfort with the idea of their child's image being stored in a digital repository, viewing it as a violation of their personal space. Others worried about the possibility of sex-selective abortions if the machine could easily determine the gender of the baby, a practice that is illegal but culturally significant in some families. These concerns highlighted that for the technology to work, the process of gathering data must be transparent, secure, and respectful of local values.
The study also uncovered a critical gap between having a diagnostic tool and having a system to act on its findings. Several health workers pointed out that even if the AI could perfectly identify a problem, the local health system might not be ready to fix it. If a machine in a small clinic detects a serious complication, but there is no ambulance to transport the patient or no specialist to treat them, the diagnosis becomes a source of fear rather than help. Participants worried that introducing advanced technology without strengthening the rest of the health system would only expose existing weaknesses. They also expressed skepticism about the longevity of such projects, fearing that the devices might arrive as short-term experiments and then disappear, leaving communities with broken equipment and no support. This sentiment, often called "pilotitis," reflected a history of well-intentioned projects that failed to become permanent parts of local healthcare.
Ultimately, the research suggests that the success of artificial intelligence in maternal health depends less on the sophistication of the code and more on the quality of the human relationships surrounding it. The people in Kenya did not reject the technology; they demanded that it be built with them, not just for them. They called for a system where the technology is transparent, where data is protected, and where the human element of care remains central. The findings indicate that for AI to truly transform maternal health in resource-limited settings, it must be co-designed with communities, integrated into a functioning health system, and guided by ethical principles that prioritize trust and equity. Without these foundations, even the most advanced tool risks becoming just another piece of hardware in a system that is already struggling to care for its people.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.