CerviScanner: AI-Assisted Cervical Cancer Screening to Bridge the Diagnostic Divide in Africa

Rwanda flag

Rwanda

Healthcare

Medium replicability and adaptation

Implementing Organisation

MednTech

Rwanda, Rwanda, Kigali

MednTech is headquartered in Ontario, Canada. The AI-assisted tool has been deployed in Rwanda.

Civil Society

Implementing Point of Contact

Maya Fakhfakh

Founder and Executive Director, MednTech

Contributor of the Impact Story

World Health Organization (WHO)

Year of implementation

2025

Problem statement

Cervical cancer remains highly preventable, yet 94% of deaths occur in low- and middle-income countries where screening quality is constrained by infrastructure and workforce limitations. In Rwanda, cervical cancer is the second most common cancer, half of cases are detected at advanced stages, and only 17.6% of healthcare providers have received proper practical screening training. Visual Inspection with Acetic Acid (VIA) is widely used because it is low-cost and enables same-day treatment, but interpretation is subjective and highly dependent on provider experience, leading to inconsistent screening quality. AI-assisted VIA tools have demonstrated strong performance in controlled research environments, but few have transitioned into routine primary care. A critical gap exists between algorithm development and real-world deployment. The challenge is not only achieving high technical accuracy but ensuring integration into existing workflows, infrastructure, regulatory systems, and provider practices within resource-constrained primary healthcare settings.

Submission Overview

MednTech is a nonprofit organization that co-designs AI-driven healthcare tools with African communities and frontline providers to expand access to preventive care in underserved settings. The organization focuses on translating AI research into deployable, context-appropriate solutions that integrate into national health systems and existing clinical workflows. MednTech emphasizes responsible AI implementation, including human oversight, regulatory compliance, data protection, and iterative improvement using local datasets. Its approach positions AI as decision-support that strengthens - rather than replaces - frontline healthcare workers, aiming to improve diagnostic consistency and health system capacity in low-resource environments.

AI Technology Used

Machine Learning
Computer Vision

Deep Learning (Convolutional Neural Network – CNN), Computer Vision for medical image analysis

Key Outcomes

Accuracy & Quality Improvement

MednTech developed and deployed an AI-assisted Visual Inspection with Acetic Acid (AI-VIA) mobile application to strengthen cervical cancer screening quality in Rwanda’s primary healthcare system. Using a convolutional neural network embedded in a smartphone-based workflow, the tool provides real-time decision support to frontline nurses during screening. In controlled validation, the model achieved 97% accuracy, and during real-world deployment at Kicukiro Health Center in Kigali, 260 women were screened with 251 complete cases analyzed, resulting in 84.5% overall accuracy under routine primary care conditions. While sensitivity decreased in field settings, the tool maintained a high negative predictive value (97.3%), reliably identifying women who do not require treatment and helping preserve limited clinical resources. Designed to augment rather than replace providers, the application integrates into Rwanda’s national VIA workflow, enhances patient engagement through image-based explanations, and demonstrates the feasibility of embedding AI decision-support tools into public health screening programs in low-resource environments.

Impact Metrics

Women screened using AI-assisted cervical cancer screening workflow

Baseline Value

NA women screened using AI-VIA tool prior to pilot deployment

Post-Implementation

260 women screened; 251 complete clinical records analyzed during pilot

Internal Monitoring·Jun 2025 - Jul 2025

Overall diagnostic accuracy of the AI tool during routine primary care screening

Baseline Value

In controlled testing conditions using a validation dataset, the AI model achieved 97% accuracy. Percentage

Post-Implementation

During real-world deployment in a public primary care facility, the AI tool correctly classified screening results 84.5% of the time when compared to expert clinical consensus. This reflects performance under routine conditions, including real patients, variable lighting, and standard workflow constraints. Percentage

Internal Monitoring·Jun 2025 - Jul 2025

Reliability in correctly identifying women who do not need treatment during screening

Baseline Value

Standard VIA screening depends heavily on provider experience, and accuracy in ruling out disease can vary across providers and facilities Percentage

Post-Implementation

During the pilot deployment, when the AI tool indicated that a woman did not have signs of cervical disease, it was correct 97.3% of the time. This means that out of 100 women identified as negative, approximately 97 truly did not require further treatment. Percentage

Internal Monitoring·Jun 2025 - Jul 2025

Implementation Context

Pilot

Kigali, Rwanda, with plans of future expansions across the Sub-Saharan African region

Women eligible for cervical cancer screening, frontline VIA nurses and primary healthcare workers, national cervical cancer program managers, and Ministry of Health stakeholders overseeing screening scale-up

Key Partnerships

Local frontline healthcare workers, Rwandan clinicians and public health researchers, the Ministry of Health, Government of Rwanda, and local research institutions

Replicability & Adaptation

Moderate

1. Performance variability between controlled and real-world settings indicates that contextual adaptation and iterative model improvement are critical for scale-up. 2. The tool may be particularly effective as a triage mechanism within HPV-based screening pathways rather than as a standalone diagnostic replacement.

Supporting Materials

* The data presented is self-reported by the respective organisations. Readers should consult the original sources for further details.