RIYADH: As Saudi Arabia accelerates its investment in AI-powered healthcare, two young researchers from the Mohamed bin Zayed University of Artificial Intelligence are building the very tools that hospitals in the Kingdom will soon need — intelligent, interpretable, and scalable systems for diagnosis and prognosis.

Although the university’s 2025 cohort did not include Saudi nationals this year, the work of two standout graduates, Mohammed Firdaus Ridzuan and Tooba Tehreem Sheikh, directly aligns with Saudi Arabia’s healthcare transformation plans under Vision 2030.

Their research offers practical, forward-looking solutions for the Kingdom’s next generation of smart hospitals.

At a time when AI systems are being deployed across diagnostic units in Saudi hospitals, from the King Faisal Specialist Hospital to new initiatives backed by the Saudi Data and AI Authority, the focus is shifting from capability to clarity.

Can the systems provide real-time support? Can they explain their reasoning? Can doctors intervene? These are the questions both Ridzuan and Sheikh have set out to answer.

Ridzuan, a PhD graduate in machine learning, developed Human-in-the-Loop for Prognosis, or HuLP for short — a cancer survival prediction system that places doctors back at the center of AI-powered decision-making.

“While AI has made significant strides in diagnosing diseases, predicting individual survival outcomes, especially in cancer, is still a challenging task,” Ridzuan told Arab News. “Our model addresses this by enabling real-time clinician intervention.”

Unlike traditional models that operate in isolation, HuLP is built for collaboration. Medical professionals can adjust and refine its predictions using their clinical expertise. These adjustments are not just temporary; they influence how the model evolves.

“Doctors and medical professionals can actively engage with the system,” Ridzuan said. “Their insights don’t just influence the result — they actually help the model learn.”

This approach to human-AI partnership ensures that predictions remain explainable, context-aware, and grounded in patient-specific realities, a key need for Saudi hospitals integrating AI at scale.

“By allowing clinicians to dynamically adjust predictions, we create a more adaptive and responsive system that can handle local challenges,” Ridzuan added.

The Kingdom’s healthcare institutions are undergoing a digital transformation driven by national entities like SDAIA, the Ministry of Health, and the Center for Artificial Intelligence in Medicine and Innovation.

These entities are focused not only on adopting new AI tools but also on ensuring that these systems can integrate into clinical workflows. This is where Ridzuan sees HuLP making an impact.

“Smart hospitals are already integrating AI diagnostic tools for medical imaging and patient data analysis,” he said. “Our model can take this to the next level by empowering clinicians to interact with and guide the system’s predictions.”

In settings where trust and transparency are vital, Ridzuan’s collaborative model could help hospitals overcome one of AI’s most persistent problems: the black box effect.

This refers to the opaque nature of certain systems, particularly in the field of AI, where the internal workings and decision-making processes are hidden or unknown.

The emphasis on local relevance also comes through in HuLP’s design. Ridzuan says real-time data from regional healthcare systems is essential for training accurate, context-sensitive models.

“Local data provides insights into the unique health conditions and medical practices within the Gulf region,” he said. “Integrating this data ensures that the AI is attuned to the specific needs and health profiles of patients in the region.

The system is built to learn continuously. As clinicians correct or refine its predictions, the model updates itself, improving with each interaction. This feedback loop is crucial for real-world deployments, especially in the Gulf, where data quality can be inconsistent.

While Ridzuan is focused on outcomes, Sheikh, an MSc graduate in computer vision, is transforming the way hospitals detect disease in the first place.

Her project, Med-YOLOWorld, is a next-generation imaging system that can read nine types of medical scans in real time.

Unlike traditional radiology AI tools, which are often limited to specific tasks, Med-YOLOWorld operates with open-vocabulary detection. That means it can identify anomalies and organ structures that it has not been explicitly trained on — a key feature for scalability.

“Most models are confined to a single modality like CT or X-ray,” Sheikh told Arab News. “Med-YOLOWorld supports nine diverse imaging types, including ultrasound, dermoscopy, microscopy, and histopathology.”

With support for up to 70 frames per second, the system is designed for clinical deployment in high-demand environments.

Sheikh sees clear potential for its use in Saudi Arabia, where institutions like the King Faisal Specialist Hospital and Research Centre are already implementing multi-modal AI imaging tools.

“It can seamlessly integrate with existing imaging systems to enable open-vocabulary detection,” she said. “Identifying a wide range of medical findings — even those outside its original training set — is essential for fast-paced clinical environments.”

But building a universal imaging tool came with its own technical hurdles.

“The biggest challenge was managing the diverse preprocessing requirements across imaging modalities,” Sheikh said. “CT and MRI scans need intensity normalization, while ultrasound, dermoscopy, and microscopy have completely different visual characteristics.”

Data imbalance was another issue. While MRI and CT scans are widely available, data for more niche imaging types is scarce. Sheikh tackled this by designing custom augmentation techniques to ensure the model performs consistently across all modalities.

She is now working on combining Med-YOLOWorld with vision-language models, systems that explain what they see in natural language.

“MiniGPT-Med does a great job at explaining radiology images,” she said. “But pairing it with a system like Med-YOLOWorld adds a crucial dimension — open-world localization. Not just describing the issue but pointing to it.”

This fusion could create a powerful end-to-end diagnostic pipeline: detect, explain, and localize. For Saudi hospitals embracing AI-driven imaging, the impact could be transformative.

For Sheikh, the global implications of her work are just as important as the technical achievements. “Med-YOLOWorld reduces the need for large, annotated datasets,” she said. “In fast-scaling healthcare systems, that’s a game-changer.”

By enabling the detection of unseen categories, the system can remain relevant even as new diseases or anomalies emerge. And when combined with language models, it can assist in medical training, annotations, and decision support, all while reducing dependence on expert-labeled data.

This approach could accelerate AI adoption in emerging regions, including across the Gulf and the wider Middle East and North Africa, where access to large datasets and AI-specialized radiologists remains limited.

While MBZUAI is based in the UAE, its alumni are playing a growing role in shaping AI initiatives that extend across the Gulf. Both Ridzuan and Sheikh have demonstrated how innovation, when aligned with clinical realities and regional goals, can scale far beyond the lab.

As Saudi Arabia continues to invest in smart hospitals, real-time imaging, and personalized care, tools like HuLP and Med-YOLOWorld represent the next wave of AI in healthcare: explainable, collaborative, and regionally adaptable.

And with growing partnerships between research institutions, healthcare providers, and government entities, these systems may not be far from deployment in the Kingdom, paving the way for a more intelligent, human-centered approach to medical care.