KBR: [Interview] Russian designer at Deep Bio overhauls AI pathology reports to drive adoption in cancer diagnostics

Why clearer, more actionable reports could drive wider AI adoption in diagnostics

If you had asked Flora Abilova 20 years ago where she saw herself, she might have pictured a courtroom, arguing cases with the sharp precision of a seasoned lawyer. Or maybe a C-suite office, a woman in a power suit making high-stakes decisions for a multinational corporation.

It definitely wouldn’t have been this.

Born in Kudymkar, a small town in Russia’s Perm Krai, Abilova pursued business management at the Azerbaijan State University of Economics. She figured influence came from mastering systems—understanding how things worked, where the power moved. By her third semester, she was already working—first at an IT company, 1CGROUP.AZ, implementing ERP solutions for government agencies and private firms.

She earned her master’s in international management at the State University of Management in Moscow, writing a thesis on how Korea supports private-sector innovation. Then—because life never goes the way you think—it was a PhD in economics at the Academy of Korean Studies.

At some point, amid all the numbers and systems, the meetings and managerial frameworks, she started noticing something: It wasn’t just the mechanics of the system that fascinated her. It was how people moved through them. What they noticed, what they ignored, what made one presentation compelling and another forgettable.

She didn’t realize it yet, but these weren’t just the questions of a strategist. They were the questions of a designer.

She’d always been drawn to design. As a kid, she sketched constantly—until one bad grade on a school assignment made her abandon drawing for years. But even as she climbed the corporate ladder, she kept coming back to it. At first, it was just a habit, then, something more.

When she started working in marketing at UNISEM in Seoul, she was suddenly in meetings with designers all the time. And there was one problem she kept running into: the visuals weren’t telling the story.

“I would explain what I wanted, lay out the concept, and the final design just—” she shook her head. “It wasn’t right. Something always felt off.”

It wasn’t that the designers weren’t good; they knew aesthetics. But they didn’t understand how an audience processed information—especially in high-stakes industries.

“I realized I didn’t just want to tell people what to design,” she said. “I wanted to do it myself.”

So she left, took formal design courses, and built a portfolio. And in September 2022, she landed her next step: Deep Bio, a Korean medical AI company developing deep-learning models to assist pathologists in diagnosing cancer.

On Feb. 5, three days after her 35th birthday, Korea Biomedical Review met with Abilova at Deep Bio’s headquarters in Seoul. She was already deep into a redesign that, as it turned out, would change everything.

Designing for trust

Deep Bio had built a powerful tool. Its AI-driven prostate cancer analysis algorithm, DeepDx Prostate, analyzes H&E-stained prostate specimens, detecting and localizing malignancies based on Gleason patterns while quantifying tumor-to-tissue ratios with 99 percent sensitivity and 97 percent specificity.

Clinically validated and CE-marked, DeepDx Prostate delivers AI-driven analysis in 30 seconds per core, enabling pathologists to rapidly assess critical areas with improved accuracy. It was also the first AI-powered pathology tool in Korea to receive Class 3 in vitro diagnostic medical device approval from the Ministry of Food and Drug Safety.

Yet, AI in diagnostics faces the challenge of interpretability. AI can detect anomalies, but it doesn’t explain its reasoning. That leaves doctors with a choice—accept the AI’s assessment at face value or ignore it altogether. Neither is ideal.

“The real challenge wasn’t just technological—it was psychological,” Abilova said. “If pathologists didn’t trust AI’s output, they wouldn’t use it.”

This is where design changes everything.

Before Deep Bio focused on usability, AI-generated pathology reports were dense and difficult to read. Critical data was buried in text-heavy layouts, making it easy to miss key findings.

“There have been cases where biopsy locations were documented incorrectly,” said Kwak Tae-yeong, Deep Bio’s chief technology officer (CTO). “When you’re dealing with prostate biopsies, where samples come from 12 different locations, a single mislabeling can lead to serious mistakes.”

For AI to succeed in clinical settings, its outputs had to be as clear and actionable as the diagnosis itself.

When Abilova joined Deep Bio, she expected her role to focus on marketing materials—brochures, posters, and exhibitions. Instead, she found herself in the middle of a much bigger issue: pathology reports that weren’t designed for real-world clinical workflows.

“These weren’t just documents. They were diagnostic tools,” she said. “And they weren’t easy to read.”

DeepDx Prostate reports contained high-value data—microscopic images, numerical breakdowns, algorithmic predictions—all critical for diagnosis. But even experienced pathologists found them overwhelming. In a high-pressure environment, excessive complexity slowed decision-making.

“I’ve seen this happen with other AI products,” Abilova said. “A great tool with poor communication leads to low adoption. I wasn’t about to let that happen here.”

She studied pathology reports from different healthcare systems. American reports used structured layouts with clear sections and bolded key points. Korean reports were dense, assuming the reader would carefully parse each detail. European reports were streamlined, stripped to essentials.

Her conclusion was clear: clarity directly impacted decision-making.

Abilova redesigned DeepDx Prostate’s reports to prioritize usability, transforming them from static documents into actionable diagnostic tools. Instead of burying key information in text-heavy layouts, the new reports used intuitive charts, color-coded overlays, and graphical indicators to make the findings immediately clear.

The feature that almost got overlooked

One of DeepDx Prostate’s strongest capabilities is its AI-generated heatmaps, which highlight potential cancerous lesions using color-coded overlays. But during early implementation, pathologists weren’t always noticing them.

Deep Bio showcased its AI-powered DeepDx Prostate reporting templates at AUA 2024. (Source: Deep Bio LinkedIn)

“I realized we had to emphasize it more,” Abilova said. “It wasn’t just another feature. It was the one that mattered.”

If the heatmaps were too subtle, doctors might miss them. If they were too bold, they could feel intrusive—like the AI was dictating a diagnosis rather than assisting in one. Finding the right balance was critical.

She refined the design, making heatmaps more prominent but not overwhelming. The impact was immediate. “Our clients found the new report format impressive,” said CTO Kwak. “They appreciated how well the information was presented at a glance.”

Abilova’s design is now fully integrated into every DeepDx Prostate report, though Deep Bio declined to share the reports, citing their proprietary nature.

From the start, Abilova said she took a deliberate approach: no exaggerated AI imagery, no sci-fi visuals, no humanoid robots.

“I never put robots in my designs,” she said. “I put people and results. I show how AI works in a way that makes sense to doctors who actually need to use it.”

The impact of AI-enhanced pathology reports extends beyond Deep Bio. Hospitals worldwide are facing a severe shortage of pathologists. In some regions, a handful of specialists were responsible for diagnosing thousands of patients each year. AI could help address these gaps—but only if clinicians feel confident using it.

“If you only have a few pathologists in an entire region,” Abilova said, “how do you prevent life-threatening delays in diagnosis?”

That’s where AI could make the biggest difference, she said. And that’s where design isn’t just about aesthetics—it’s about clinical impact. Because in the end, the problem isn’t whether AI can detect cancer. The problem is whether doctors can trust what they see.

“There’s this assumption that AI is this big, cold, robotic thing,” Abilova said. “But at the end of the day, what we’re trying to do is make life easier for humans. And I think that’s the part people forget sometimes.”

SOURCE: Korea Biomedical Review