At Singapore’s Changi General Hospital (CGH), artificial intelligence isn’t treated as a buzzword or an experimental add-on. It is embedded into operational thinking, governance structures, and care delivery. Much of that momentum can be traced to Dr Charlene Liew, Senior Consultant, Department of Diagnostic Radiology and Director, AI Office at CGH, who oversees more than 70 AI projects spanning ideation to deployment.

 

Six of these are already in clinical use.

“AI must be measurable, safe, and meaningful,” she says. “Our role isn’t to chase trends, it’s to ensure AI improves care, service delivery, and patient outcomes.”

Her approach reflects a maturing era of healthcare AI: disciplined governance, strong cross-functional alignment, and a willingness to confront the uncomfortable truth that innovation must deliver real value, not just novelty.

From Ideas to Impact: How CGH Chooses the Right AI Projects

Charlene is the co-chair of the hospital’s AI & Digital Committee, a cross-functional body covering finance, pharmacy, allied health, clinical operations, and digital services. The committee meets regularly to evaluate and prioritise proposals, a process she describes as “pitch-style”.

“We put the problem statement at the centre,” she explains. “If a department can articulate the operational gap, we help them find the right solution whether internal, external, or hybrid.”

Rather than letting vendors dictate direction, CGH runs a clearing-house model: matching needs to innovators while ensuring governance, feasibility, and alignment with strategic priorities. This prevents duplication, accelerates promising ideas, and ensures scarce resources go to high-value areas.

A Case Study in Adoption: AI Triaging in Radiology

One of CGH’s most successful deployments is an AI triage tool for radiology X-rays. The system flags critical, urgent, and normal cases, helping radiologists prioritise reporting more efficiently.

Since implementation, the smart solution has reduced the average turnaround time for inpatient cases by up to 97%. The system now triages 100% of urgent cases and has reduced urgent case reporting turnaround times by up to 50%s. Importantly, Charlene notes; the tool has faced little clinician pushback.

“Because it doesn’t replace clinicians,” she says. “It supports them in making faster, more accurate decisions. That distinction matters.”

Where AI Needs to Grow: Operational Innovation

While radiology dominates the global AI landscape, Charlene notes that 75% of FDA-approved tools are radiology-related. She argues that the next frontier is operational AI.

“Scheduling, workload balancing, resource planning are areas that aren’t glamorous, but they’re where AI can have enormous system-level impact.”

Today, radiology scheduling at many hospitals continues to be manually performed. Workload distribution is often inequitable because it doesn’t account for case complexity. These operational frictions compound pressure on clinicians and slow care delivery.

“We need AI to improve the core parts of healthcare,” she says. “That’s where the real transformation will happen.”

The ROI Blind Spot: Rethinking How We Measure AI Value

One of the biggest challenges in AI adoption, Charlene argues, is financial justification. Many AI benefits avoided ICU admissions, faster radiology turnarounds, timely interventions as they don’t fit neatly into traditional ROI calculations.

“How do you monetise the prevention of deterioration?” she asks. “Traditional metrics weren’t built for these types of outcomes.”

She warns that if budget scrutiny tightens while metrics remain outdated, the sector may enter a period where innovation slows because evidence frameworks cannot capture AI’s true value.

“We need modernised impact assessment tools. Otherwise, we risk undervaluing what AI brings to care.”

A Data Tsunami Is Coming — and Most Institutions Aren’t Ready

Charlene is also sounding the alarm on a challenge few organisations are fully prepared for: exponential growth in healthcare data.

Imaging, electronic health records, genomics, and multimodal inputs are expanding rapidly. With that comes escalating storage requirements, rising costs, and heightened cybersecurity risks.

“Most hospitals underestimate what this means operationally,” she says. “We need long-term infrastructure planning, not incremental fixes, if we want AI to scale safely.”