By Nwafor Sunday

Nigerian health technology professional, Oyindamola Adebayo, has developed a framework aimed at improving the reliability and safety of artificial intelligence (AI) tools used in medical imaging across African healthcare facilities.

The framework, known as the Quality-Control and Multimodal Clinical Intelligence (QC-MCI), seeks to address challenges associated with AI-assisted radiology, including poor image quality, technical errors, incomplete patient information and unreliable internet connectivity.

Adebayo said the framework was designed to ensure that AI-generated medical assessments were based on reliable imaging data and relevant clinical information before being used to support medical decisions.

It evaluates image quality, identifies possible technical defects and integrates patient information such as symptoms, medical history, laboratory results and previous medical examinations.

The framework also provides for cases to be referred to medical professionals for further assessment where clinical information is incomplete or an AI system produces results with low confidence.

Adebayo said the effectiveness of AI in radiology should not be judged solely by the accuracy of its algorithms but also by how reliably the technology performs in everyday healthcare settings.

“The central implementation problem is no longer whether an algorithm can produce a strong score on a retrospective test set. It is whether the combined system; data, software, radiologist, workflow and governance continues to produce safe and useful decisions when it encounters routine clinical variation,” she said.

She said her work also involved reviewing proposed technologies through professional channels associated with radiography practice in Nigeria.

According to her, such assessments consider technical performance, quality-control procedures, operational reliability, implementation requirements and compatibility with existing healthcare infrastructure.

She noted that healthcare facilities across Sub-Saharan Africa operate under varying conditions, including differences in medical equipment, internet connectivity, availability of specialists and financial resources.

She said QC-MCI was developed to accommodate these differences rather than impose a uniform implementation model on all healthcare facilities.

The framework allows individual hospitals and diagnostic centres to adapt image-quality checks, clinical information integration, human supervision and referral procedures to their operational capacity.

Adebayo recommended a phased implementation approach, beginning with the verification of medical images before integrating imaging results with relevant patient information.

She also advocated testing AI systems across different patient groups and healthcare facilities, introducing AI-generated results at controlled stages of radiologists’ workflows and documenting disagreements between medical professionals and automated systems.

She said systems should be suspended or recalibrated whenever monitoring identifies potential safety concerns.

The framework comprises five major functions: input quality control, multimodal clinical context integration, model and subgroup validation, human-AI interaction control, and continuous post-deployment monitoring with escalation procedures.

According to Adebayo, the components are intended to ensure that AI tools remain dependable after deployment, particularly when exposed to varying clinical conditions and patient populations.

The framework is also designed to help healthcare providers identify inadequate medical images early, reduce unnecessary repeat examinations, flag cases requiring specialist attention and provide clinicians with relevant information for informed decisions.

Adebayo said the growing adoption of AI in African healthcare required greater attention to the processes surrounding automated medical assessments, rather than relying exclusively on performance claims by technology developers.

She maintained that adapting AI systems to different healthcare environments while retaining quality controls and professional oversight would be critical to their effective use.

She said QC-MCI was intended to promote accountability in AI-assisted radiology, with technology complementing, rather than replacing, the clinical judgement of medical professionals.