Can AI become a trusted frontline diagnostician?
In communities across Zimbabwe, AI-powered tools are helping frontline workers detect disease earlier and facilitate timely care. Scaling these technologies will depend on ensuring accuracy, building trust, and strengthening infrastructure.
A digital health promoter using a device to measure Janet Museta's blood pressure near Nyanga, Zimbabwe. Photo: Farai Shawn Matiashe
When Janet Museta woke up in a hospital in Zimbabwe’s Manicaland province, after being unconscious for hours, she was not as much worried about the extent of her injuries as about the risks of medical emergencies like stroke and high blood pressure that would impede her ability to work and take care of her family.
The 60-year-old mother of five had lost a leg after being hit by a speeding motorcycle at a shopping centre in her home village of Gwanyamwanya outside Nyanga, a popular tourist town north of Zimbabwe’s eastern highlands.
After surgery in February, Museta learned how to use her new artificial limb. But after a month, her health took a turn for the worse. She started feeling dizzy and experiencing severe chest pain.
A digital health promoter—a community health worker who uses digital tools to provide basic screenings and health education—came to her house to measure her blood pressure, revealing that it was extremely high. Museta spoke to a doctor through a video call on a smartphone, who diagnosed her with hypertension and prescribed medication, which Museta takes once a day.
“I believe my blood pressure was high because of stress. It has not been easy,” says Museta, who survives on farming. “Imagine waking up one day, realising that you cannot do household chores or go to the farm.”
Alarmingly, she then started experiencing pain on the right side of her abdomen and numbness—warning signs of an imminent stroke. The digital health promoter used a digital tool called Afya Selfie to assess her stroke risk. Thankfully, it was found to be minimal. “I was relieved. I feared the worst,” Museta says.
Afya Selfie is a Zimbabwean AI screening tool that uses a smartphone camera and photoplethysmography—a technique that measures blood volume changes using light—to estimate indicators such as blood pressure, heart rate, and stress levels. Through a selfie scan, the app assesses cardiovascular risk. Patients pay between US$1–3 for screening.
It is one of a growing number of AI-powered technologies that are expanding digital access to health assessments across Zimbabwe and beyond, bringing speedy diagnosis and specialist support to underserved communities.
AI-enabled diagnostic tools
In March, Zimbabwe launched its National Artificial Intelligence Strategy (2026-30), aimed at accelerating the adoption of AI in different sectors, including health. Against this backdrop, innovators have been deploying a growing number of AI tools such as Afya Selfie to serve as diagnostic aids, particularly in rural areas.
Dr Admore Jokwiro, a co-founder and chief medical officer at ZimSmart Villages, which developed Afya Selfie, says the AI tool works by combining photoplethysmography, smartphone-camera imaging, machine-learning models trained on large clinical datasets, and conversational AI interfaces to assess a patient’s risk profile in real time.
“Together, these tools function as a…first line of clinical intelligence that can triage, flag, and refer patients who would otherwise go undetected for years,” he says.
AI smartphone screening tool Afya Selfie assesses Janet Museta's risk of having a stroke near Nyanga, Zimbabwe. Photo: Farai Shawn Matiashe
Zimbabweans have been grappling with an underfunded health system for decades. There are shortages of not only basic medicine like paracetamol but also of health workers, including doctors, specialists, and nurses. The ratio of physicians to the population is critically low; specialist services are concentrated in urban centres, and rural and semi-urban communities face a rising burden of non-communicable diseases like cardiovascular disease, diabetes, hypertension, anaemia, and different types of cancer.
“AI tools address that gap directly by bringing screening to the community rather than waiting for the community to reach a clinic,” Jokwiro explains.
To date, nurses and digital health promoters have used Afya Selfie’s cardiovascular risk predictor to screen more than 5,000 people, with a striking 30–35% returning abnormal results like high hypertension or cardiovascular disease risk. The results show that for every three people screened, at least one is carrying a metabolic or cardiovascular risk factor that was previously invisible to the health system, says Jokwiro.
ZimSmart Villages has since expanded its suite of AI-powered screening tools. Its breast cancer detection app uses image analysis algorithms to identify suspicious lesions from photographs taken with a handheld ultrasound scan. Using the technology, experts have assessed more than 3,000 women, detecting suspicious findings in around 10% of cases and referring them for further investigation.
Another tool, Anemia Checker, estimates haemoglobin levels non-invasively by analysing images of the lower part of the eye with a smartphone, offering an alternative to laboratory-based blood testing. Project Tendesai, meanwhile, screens for depression, trauma, and substance use disorders using validated psychometric assessments delivered in local languages.
Speed and accuracy of AI tools
Professor Tawanda Mushiri, executive director at the Scientific & Industrial Research & Development Centre, which provides technological solutions for sustainable development in Zimbabwe, says that although the accuracy and reliability of AI in diagnostic tools vary enormously, there are cases where they have outperformed traditional methods of providing healthcare.
One example, he says, is the non-invasive anaemia detection application (NiADA) developed by Dawa Health, a digital health provider operating in Zimbabwe, Rwanda, and Zambia. The AI-powered tool estimates haemoglobin levels without requiring a blood test, enabling healthcare workers to identify anaemia in pregnant women before childbirth and helping prevent potentially life-threatening complications. “Dawa Health’s tools reportedly cut anaemia detection from three days to five minutes in rural pilots,” Mushiri says.
A non-invasive AI tool, Anemia Checker, estimates haemoglobin levels without requiring a blood test by analysing images of the lower part of the eye with a smartphone. Photo: Hanna Saad
Jokwiro says his organisation’s screening tools have been validated against established clinical reference standards and have demonstrated strong performance. He says the tools correctly identify people with a health condition and accurately rule out those without one in more than 85% of cases. According to him, this level of accuracy is comparable to the clinical standards routinely achieved by diagnostic tools used in well-equipped hospitals in high-income countries.
“Sensitivity above 85% means the tool correctly identifies the vast majority of individuals who truly have the condition being screened for, minimising the risk of false reassurance,” Jokwiro points out.
Building trust in AI as a healthcare assistant
Securing the trust of health workers and patients in AI’s capabilities is often just as important as developing the technology itself. When Museta was told that the risk of having a stroke in the future was minimal, she was sceptical and struggled to believe a smartphone could provide such a reliable assessment. “I asked the health worker a lot of questions and her answers were convincing,” Museta says.
Annamore Murenzvi, from Hauna, a farming community next to Nyanga, came for a pregnancy ultrasound scan in October last year. Using a tablet’s camera, an AI-powered tele-obstetrics app developed by ZimSmart Villages detected twins and that she was already in her second trimester. Health workers monitored her until she gave birth to two bouncing baby boys early this year. “I did not have any complications. I have always had trust issues with scans, but this one was accurate,” says the first-time mother.
“I have confidence in this AI tool. It helps us check the health of both the mother and the foetus. It reduces anxiety for the mother.”
Using the app, nurses perform obstetric scans using portable devices, supported by remote specialists. The app tracks gestational progress, predicts preeclampsia risk, and monitors foetal wellbeing.
Memory Difara, a registered nurse and midwife at Zimsmart Villages’ telehealth centre, says that when Murenzvi came for the scan, the app did not detect any abnormalities.
“The foetal heart rate was normal. We screened, using the app, for major birth defects like anencephaly, cleft lip, and spina bifida, which were normal. The placenta was healthy, umbilical cord and amniotic fluid were normal,” she says. “The AI app calculated the gestational age and estimated foetal weight and date of delivery. This helped the mother to plan for the delivery of her babies.”
An AI-powered tele-obstetrics app was used to monitor Annamore Murenzvi's pregnancy for her twins in Huana, Zimbabwe. Photo: Farai Shawn Matiashe
Difara believes the diagnostic tool is accurate. “I have confidence in this AI tool. It helps us check the health of both the mother and the foetus. It reduces anxiety for the mother,” she says.
Jokwiro emphasises that the AI tools are designed as screening instruments, not definitive diagnostic replacements, explaining that their role is to identify people who need further investigation, rather than replace a clinician’s final judgment.
Mushiri agrees that the tools work best as decision support, not replacement. “The framing that keeps recurring in the literature is AI as an assistant to a health worker, not a substitute for one. AI-driven decision-support systems can act as intelligent assistants that enhance clinical reasoning, promote adherence to guidelines, and improve accuracy, rather than [replace] healthcare professionals,” he says.
Towards a long-term vision
Despite their potential, integrating these AI tools into Zimbabwe’s health system at scale remains challenging. Jokwiro notes that connectivity and device access remain a tough call. He says while smartphone penetration is growing rapidly, the most remote and vulnerable communities still lack reliable internet access and affordable devices.
Mushiri says regulatory and accountability gaps persist. “It is unclear how the Medical and Dental Practitioners Council of Zimbabwe can regulate an AI-based app that diagnoses and issues prescriptions,” he says, noting there is no clear regulatory and liability framework.
“A genuine AI health infrastructure for diagnostics in Zimbabwe and beyond will rely on open standards, locally owned data infrastructure, and a training pipeline.”
Mushiri adds that while trust is a known adoption barrier, workforce training gaps make it worse. “Healthcare workers often lack training in interpreting algorithmic outputs or integrating machine-learning insights into clinical practice,” he says, explaining that health workers operating a genuinely accurate tool can misuse it simply because no one trained them on what the output means.
Jokwiro, as a health data expert, believes that a genuine AI health infrastructure for diagnostics in Zimbabwe and beyond will rely on open standards, locally owned data infrastructure, and a training pipeline. This pipeline will continuously improve performance by using African patient data.
Mushiri says a practical starting point is acknowledging that infrastructure, not algorithms, is the bottleneck. He highlights that there is a need to enhance devices and connectivity first and implement low-data, offline, and localised models.
The practicality of implementing AI-powered tools must start by addressing the infrastructure and need for low-data, offline, and localised models. Photo: Iwaria Inc.
To keep costs down, tools such as Afya Selfie and the tele-obstetrics app are deployed through existing infrastructure, with ZimSmart Villages transforming 12 state-owned post offices into telehealth centres across the country. Nurses based at these centres use AI-powered tools to screen patients, while people who need further care can consult specialists remotely through telehealth services. “In the immediate term, the priority is scaling the post office clinic model,” Jokwiro says, noting that there are more than 200 post offices across Zimbabwe that could be used.
He says the longer-term vision is a national AI health screening network that feeds anonymised, aggregated, epidemiological data into the Ministry of Health’s planning systems, giving policymakers, for the first time, a real-time picture of disease burden at the community level rather than relying on hospital admission data, which captures only the most severe cases.
For patients like Museta, the promise of this technology is not just better data for policymakers, but earlier detection and reassurance at a time when uncertainty can be frightening. “I thought I was dying, not of the injuries but of a stroke because of stress. I could not sleep. But I am feeling better now,” she says.
