As Healthcare AI Matures, Trust Is Expanding Beyond the Model How Trust in Healthcare AI Is Expanding Beyond Model Performance

As healthcare AI adoption expands globally, the way AI companies are evaluated is also beginning to change. Until recently, conversations around healthcare AI were primarily centered on model performance, including sensitivity, specificity, clinical validation, and regulatory approval. These factors continue to matter deeply.

But as AI systems become more integrated into real clinical environments and workflows, another layer of evaluation is emerging alongside technical performance.

Healthcare institutions, pharmaceutical companies, and global partners are increasingly assessing not only whether an AI model performs well, but also whether the organization behind it can operate as a reliable long-term partner within complex healthcare ecosystems.

That shift is gradually changing what “trust” means in healthcare AI.

Expanding Expectations Around Healthcare AI

Healthcare AI is no longer being deployed only as standalone pilot technology.

Today, AI systems are becoming part of broader clinical workflows, diagnostic infrastructure, and operational decision-making processes.

As adoption expands, healthcare organizations are beginning to evaluate not only whether an AI model performs well, but also whether the company behind it can support stable and responsible operations over time.

This includes areas such as governance structures, transparency, risk management processes, sustainability practices, and organizational accountability.

As Lunit’s IR&G team explains, healthcare AI companies are increasingly being evaluated not only for their technology, but also for “whether the company can develop and operate that technology in a trustworthy and responsible way.”

In other words, trust in healthcare AI is gradually extending beyond the model itself.

ESG and Long-Term Healthcare Partnerships

ESG (Environmental, Social and Governance) considerations are becoming increasingly visible in how global healthcare partnerships are formed. In many industries, ESG evaluations were once viewed primarily through the lens of corporate reputation or investor communications.

But increasingly, global organizations are incorporating ESG-related requirements and sustainability assessments into supplier reviews and partnership processes. Healthcare is no exception.

According to Lunit’s IR&G team, the global healthcare industry is also placing growing importance on operational areas such as ethics and compliance, information security, quality management, human rights, and environmental management systems, alongside technological capabilities.

As healthcare AI companies work more closely with hospitals, pharmaceutical companies, and public healthcare systems, expectations around operational maturity are becoming more structured.

Long-term collaboration now often involves broader discussions around governance, compliance, operational processes, and organizational resilience.

From this perspective, ESG is becoming less about messaging and

more about how companies demonstrate operational readiness.

Building a Global Sustainability Management Framework at Lunit

Against this backdrop, Lunit has continued strengthening its ESG practices and broader sustainability management framework in line with global standards.

Recently, Lunit received the Silver rating from EcoVadis, one of the world’s leading sustainability assessment platforms, for the second consecutive year. Lunit achieved a total score of 76 out of 100, placing the company in the top 9% of companies assessed by EcoVadis. This marked a significant increase from last year’s score of 69.

The company received particularly strong scores in Environment (84) and Ethics (80).

Lunit also received strong recognition for its environmental policies and global initiatives, as well as its ethics policies.

Lunit has committed to the Science Based Targets initiative (SBTi) and is in the process of seeking validation for its near-term Scope 1, 2, and 3 targets. Separately, the company has set an internal target to achieve net zero Scope 1 and 2 emissions by 2040. The company has also joined the United Nations Global Compact (UNGC), expanded renewable energy certificate (REC) purchases, and continued strengthening its information security, ethics, and compliance practices.

In addition, Lunit has continued to strengthen people-related practices that support responsible and sustainable growth, including employee development, regular performance reviews, employee engagement programs, and family-friendly workplace policies.

These efforts are not just about external ESG messaging or assessment results. Rather, they reflect a broader effort to build operational trust and organizational readiness increasingly expected within the global healthcare AI industry.

As the Healthcare AI Industry Evolves

The healthcare AI industry itself is also entering a different stage of maturity.

In the earlier phases of adoption, the primary focus was naturally on proving whether AI technology could deliver meaningful clinical performance.

Today, however, conversations are evolving beyond technical feasibility alone.

Questions around scalability, integration, accountability, and long term operational trust are becoming increasingly important as healthcare AI moves deeper into real world healthcare systems.

Lunit’s IR&G team describes this transition as a shift from evaluating “AI model performance alone” toward assessing whether companies can develop and operate AI “stably and responsibly” across broader healthcare environments.

This reflects a broader transition within the healthcare AI industry. Healthcare AI is no longer evaluated only by what the model can achieve, but also by how responsibly and reliably the organization behind it can operate.

Trust as Part of Healthcare AI Operations

As healthcare AI becomes more deeply embedded into healthcare systems globally, operational trust is also becoming an increasingly important part of the conversation.

Performance will always remain fundamental.

But in healthcare, where technologies directly interact with clinical workflows, patient outcomes, and institutional decision making, trust is built through a much broader operational foundation.

This is why ESG practices and sustainability management frameworks are increasingly being viewed not simply as external evaluation tools, but as part of the infrastructure supporting long term global partnerships.

Lunit aims to continue strengthening not only its AI technologies but also the operational systems and organizational foundations required to become a trusted long-term partner within the global healthcare ecosystem.

The healthcare AI industry is gradually moving in that direction.

Why Healthcare AI Works Differently in Japan? Why Lunit chose to build a local presence and what operating in Japan reveals about medical AI in practice

In Japan, discussions around the real-world application of healthcare AI have been steadily gaining momentum.

The Ministry of Economy, Trade and Industry recently hosted a SaMD Forum, bringing together global medical AI companies to share their experiences and challenges in commercialization.

Lunit was invited to take part in this conversation,
offering perspectives shaped by its experience operating in the Japanese market.

These developments suggest that Japan is moving beyond evaluating technology and is increasingly focused on how healthcare AI can be applied in practice.

Against this backdrop, Lunit established its Japan subsidiary in May 2025.
Rather than functioning as a conventional sales office,
the entity was designed with a broader role in mind,
serving as a direct hub connecting the Asia-Pacific region.

What kind of market is Japan

Japan appears to be a market where healthcare AI meets real-world conditions relatively early.

An aging population, workforce constraints,
and an imaging-heavy clinical environment
naturally create space for AI to play a meaningful role.

At the same time, the level of social acceptance toward medical AI
seems comparatively high.

In some areas, reimbursement for AI-based medical technologies is already being discussed and applied.

Taken together, these factors make Japan feel less like a market of potential and more like one where practical adoption is beginning to take shape.

Why Japan is a difficult market to enter

At the same time, Japan remains a demanding market.

The PMDA approval process requires rigorous validation,
including local clinical evidence and careful localization.

Aligning with existing clinical workflows
and building trust with customers
are also processes that tend to take time.

As Kyungsik Jo, Head of Lunit Japan Inc., explains,

“Japan is a highly important market, with large scale and very high standards across purchasing criteria, regulatory requirements, and clinical validation.
That also means the barriers to entry are high for global companies.”

One of the most significant challenges was the PMDA approval process.

“Even though we already had sufficient clinical evidence and commercial experience globally, entering Japan required a separate regulatory and clinical strategy tailored to local standards. We had to carefully align on clinical protocols, primary endpoints, and sample sizes, and even coordinating meetings and receiving feedback took considerable time.”

Ultimately, the key realization was this:

“In Japan, it is not enough to say a product has already been validated elsewhere. It needs to be re-explained and re-proven in a way that aligns with local expectations.”

This reflects how the market operates under a distinct set of conditions.

Where Product–Market Fit Becomes Visible

Despite these challenges, Japan has become one of Lunit’s most important markets.

Lunit INSIGHT currently records its highest revenue in Japan,
suggesting a strong alignment between the product and local needs.

High imaging volumes and strong expectations around accuracy
make the value of AI-assisted interpretation more visible in everyday practice.

As Kihwan Kim, Head of CSG at Lunit, noted,

“Japan is a market where we were able to achieve strong product-market fit for AI-powered diagnostic support. Given the shortage of specialists and the prevalence of non-specialists interpreting medical images, we believe Lunit’s technology can clearly contribute to improving diagnostic performance.”

In this sense, Japan can be seen as an environment
where product-market fit becomes easier to observe.

From partnership to direct engagement

Lunit initially entered Japan through a partnership with Fujifilm,
leveraging its established distribution network.

This approach played an important role in early market entry.
At the same time, it gradually became clear
that a deeper understanding of the market would require a different kind of engagement.

As described by Kyungsik Jo,

“Lunit’s strategy in Japan can be summarized as building on strong partnerships while strengthening direct engagement where needed.
This is not about competing with Fujifilm,
but about combining the strengths of both organizations
to better cover the Japanese market.”

Following the establishment of its local entity,
Lunit has been expanding toward a hybrid approach
that combines partnerships with direct engagement.

Rather than simply adding another sales channel,
this shift reflects an effort to better understand how the market actually works.

Direct interaction allows for a closer look at clinical workflows,
decision-making processes, and the constraints that shape real-world adoption.

Japan as a benchmark for APAC strategy

Lunit does not view Japan simply as another overseas market.

Instead, it is increasingly seen as a reference point
for the company’s broader Asia-Pacific strategy.

In Japan, every step of the process
from regulatory approval and clinical validation
to partnerships and customer engagement
tends to be held to a higher standard.

As Kyungsik Jo explains,

“Japan and Korea are among the most advanced healthcare markets in Asia.
The clinical evidence and execution experience we build in these markets
go beyond regional success and contribute to strengthening trust in medical AI across APAC. We see growth in Japan not just as revenue expansion, but as raising the bar for our overall regional strategy.”

Experience built in this environment often carries over into other markets,
shaping how strategies are designed and executed elsewhere.

Closing

One thing seems to be becoming clearer over time.

The question around healthcare AI is gradually shifting
from what is possible to what can be sustained in real-world environments.

And that answer does not emerge from a single market,
but from the accumulation of experience across different ones.

In that sense, Japan is less a destination and more a point of reference.

As Kyungsik Jo puts it,

“The Japan subsidiary is a highly strategic organization.
It is not just a local sales office, but a key hub that translates Lunit’s global capabilities into real outcomes in the Japanese market and feeds those learnings back into our broader APAC and global strategy. Success in Japan will not only drive local growth, but also serve as a meaningful milestone for Lunit’s global expansion.”

For Lunit, the experience built here is not only shaping how the company operates locally, but also defining how it moves forward globally.

How AI is moving into real-world use and what Lunit has seen along the way

Why the world gathers in Davos

Every January, the world gathers in Davos.

The 2026 World Economic Forum (WEF) was no exception.
Leaders from politics, technology, and healthcare came together in one place from heads of state to AI pioneers, as well as executives from leading pharmaceutical companies, including figures such as Donald Trump, Elon Musk, and Bill Gates.

Davos is more than a conference.
It is where the direction of industries and future ecosystems begins to align.

What WEF revealed about AI

At this year’s WEF,
several consistent themes around AI emerged.

First, the conversation has shifted
from whether to adopt AI
to how to apply it across organizations.

Second, the impact of AI is no longer framed primarily as job displacement,
but as a transformation of roles and required capabilities.

Third, as adoption accelerates,
responsible design and operation are becoming critical priorities.

Why AI is harder in healthcare

These dynamics do not apply equally across industries.
Healthcare remains one of the most complex environments for AI.

Regulation, clinical validation, integration with existing systems,
and trust from both practitioners and patients
must all be addressed at once.

As a result, AI in healthcare is not simply about adoption.
It requires time, evidence, and sustained relationships.

What Lunit has seen at Davos

The World Economic Forum (WEF)
is where many of these relationships begin.

Lunit has attended WEF for four consecutive years,
participating as the only medical AI company on a global stage.

Meaningful partnerships are rarely formed in a single meeting.

The first interaction builds context.
By the second or third,
the conversation becomes more concrete.

At WEF 2026,
CEO Brandon Suh met again with the Minister of Health of a major European nation,
following their initial introduction the year before.

This is how relationships begin to move toward collaboration.

How relationships turn into real-world impact

This pattern is reflected in Lunit’s partnerships.

The collaboration with AstraZeneca
originated from a meeting at WEF
and later developed into an ongoing business relationship.

Similarly, relationships with IQVIA, Microsoft, and Agilent
have either begun or deepened through interactions at WEF.

In public health,
initial exchanges have evolved over time
into the deployment of AI-based diagnostic solutions
within real healthcare systems.

Looking Ahead

WEF is not just a global event.
It is where relationships begin and evolve into real collaboration.

The same dynamic applies to AI.

As AI moves from adoption to application,
the defining factor is no longer the technology itself,
but the ability to implement it in real-world environments.

In this phase, competitive advantage is shaped not only by speed,
but by the ability to navigate complexity, build trust,
and sustain relationships over time.

Within this shift, Lunit has focused not only on developing AI,
but on applying and scaling it in real clinical settings.

Through global partnerships and continued engagement at WEF,
relationships built over time have translated into real business outcomes
reflecting how this transition takes shape in practice.

As AI continues to evolve,
Lunit is building this experience
to help shape how AI is applied in real-world healthcare.

What Is AI in Pathology Really Changing?
Changho Ahn, Head of Medicine, Oncology Group at Lunit, on AI’s Expanding Role in Clinical Decision-Making

Introduction | Treatment Has Become More Precise, but Choices Have Not

Cancer treatment has advanced rapidly.
Yet as treatment options continue to expand, clinical decision-making has not become simpler — if anything, it has grown more complex.

From immuno-oncology to targeted therapies and ADCs, physicians today are no longer asking whether a treatment exists, but which treatment is right for this patient, right now.

Pathology sits at the starting point of that question.
And as AI enters pathology, its role is shifting — from the final step of diagnosis toward a place where treatment choices begin to take shape.

This article is based on an interview with Changho Ahn, Head of Medicine, Oncology Group at Lunit, and examines how AI-driven pathology is being applied across clinical practice and drug development.

How Pathology Is Becoming Data-Driven

Ahn describes pathology today as a field still in transition.

“In many hospitals, pathologists are still reading slides through microscopes.
Some large institutions have introduced digital pathology, but overall,
the transition is still underway.”

Within this environment, AI-driven pathology is introducing three fundamental changes.

• First, quantification.
  AI analyzes every cell on a slide and replaces qualitative impressions — such as “high” or “low” immune infiltration — with precise measurements. These quantitative insights can be used as biomarkers to help predict treatment response.
• Second, reproducibility.
  Where results once varied by reader, AI applies the same criteria consistently. This consistency is especially important in areas like PD-L1 expression testing, where small thresholds carry significant clinical weight.
• Third, pattern discovery.
  Treatment response is increasingly linked not only to how much a target is expressed, but where and how it is distributed within tissue. AI can quantify these spatial patterns, capturing signals that are difficult to assess visually.

Together, these changes signal a broader shift: pathology is becoming actionable information for treatment selection.

Biomarkers in an Era of Increasingly Difficult Choices

Despite major advances, treatment decisions in late-stage cancer remain challenging.

Evaluating whether a therapy is effective can take months, and response rates often remain below 50%. Patients, however, rarely have that time.

Ahn explains the dilemma from a clinical perspective:

“Cancer treatment has advanced, but for late-stage patients, cure is still difficult. Survival is often around a year, yet it can take months just to determine whether a treatment is working.”

In this context, biomarkers play an increasingly critical role. By identifying patients more likely to respond, biomarkers reduce unnecessary treatment burden and allow physicians to make more confident choices.

Their value extends beyond the clinic. For pharmaceutical companies, better patient stratification can improve trial success rates, shorten development timelines, and increase the likelihood of regulatory approval.

Why Lunit Chose Pathology

Lunit operates under a clear mission: to conquer cancer through AI. That mission has taken shape along two strategic paths.

“One is finding cancer faster and more accurately.
The other is helping ensure cancer is treated more effectively.”

While Lunit does not develop drugs, it identified biomarker-driven patient selection as an area where AI could deliver meaningful impact.

Medical imaging and pathology both rely on complex visual data — an area where AI excels. Within this context, pathology emerged as a central pillar of the Lunit SCOPE strategy.

Performance Validated Through Competition

The most credible way to evaluate pathology AI is through head-to-head comparison under identical conditions.

Global pharmaceutical companies often provide the same datasets to multiple AI vendors, withhold ground truth, and select partners based solely on results.

Ahn describes the process:

“Pharmaceutical companies evaluate AI vendors almost like an exam.
They give the same data to multiple companies and judge only by the results.”

Through repeated participation in these evaluations, Lunit has demonstrated strong performance, leading to continued project selection and collaboration. One such evaluation marked the starting point of Lunit’s partnership with AstraZeneca.

How Data Drives Model Advancement

At the core of AI model development lies data — both large-scale image datasets and high-quality annotation.

Lunit initially partnered with academic hospitals and later expanded through strategic collaborations to build large pathology image repositories. Annotation — labeling individual cells as immune or cancer cells — is performed with direct involvement from Lunit’s internal pathologists and external experts.

Ahn emphasizes why this matters:

“In pathology AI, data quality matters more than sheer volume.
Who annotated the data, under what criteria, and how consistently
that defines how the model learns.”

Lunit also employs a human-in-the-loop approach, allowing models to improve continuously through expert feedback. More recently, the company has been developing foundation models designed to support a wide range of pathology tasks.

Beyond Pathology: The Direction of Lunit SCOPE

Lunit’s goal is not simply to build AI software, but to deepen understanding of cancer and support better clinical decision-making.

Digital pathology is a powerful starting point, but not the only one. When combined with CT, MRI, and other clinical data, pathology contributes to a more comprehensive, multimodal view of disease.

Lunit has already presented AI models predicting immunotherapy response from lung CT scans at ASCO. Looking ahead, multimodal AI — integrating pathology, imaging, and additional data — remains a key direction.

Conclusion | Pathology as the Beginning of Decisions

AI-driven pathology is not about replacing pathologists. It is about establishing clearer standards through quantification, ensuring consistency through reproducibility, and uncovering patterns that support better treatment choices.

Ahn summarizes the focus of Lunit SCOPE:

“At Lunit, we are working with global pharmaceutical partners to develop new biomarkers, expanding beyond immuno-oncology into antibody-based therapies, and preparing for multimodal expansion. Guided by Lunit’s mission to conquer cancer, we continue to move forward — one step at a time.”

A National Project to Build a
Medical Science Foundation Model And Why Lunit Was Chosen to Lead It

What makes this initiative notable is not a single AI model, but the effort to build shared AI infrastructure that connects the entire medical science lifecycle at a national scale.

Lunit has been selected as the lead organization and overall technical and clinical validation coordinator for South Korea’s national initiative to develop a medical science–specialized AI foundation model, led by the Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA).

Rather than a conventional R&D program, the initiative reflects a broader national strategy: to connect the full medical science lifecycle, from molecular research to real-world clinical decision-making, within a single AI system.

Why a Full-Cycle Medical Science Foundation Model Now

For decades, medical science data has remained fragmented across separate systems. Molecular and omics data, drug information, medical literature, clinical guidelines, and hospital-generated clinical data have typically been managed in isolation.

This fragmentation has made it structurally difficult to translate research insights into clinical impact, or to feed real-world clinical outcomes back into research.

Against this backdrop, the initiative addresses several long-standing challenges in medical science AI:

• Fragmentation between research data, clinical evidence, and real-world outcomes
• Limited feedback loops between clinical practice and scientific discovery
• Difficulty scaling validated medical AI across healthcare systems

To address these challenges, the project brings together the full medical science lifecycle — spanning molecules and omics, drugs, literature, guidelines, clinical data, and real-world hospital validation — within a single foundation model.

Designed as a shared AI base, this foundation model supports multiple downstream applications across healthcare and life sciences. Importantly, validation across 11 hospitals participating in the Lunit-led consortium allows the model to complement theoretical performance with evaluation in real-world clinical environments.

“There are scientists who specialize in individual domains, but it is rare to find medical scientists with expertise across the entire spectrum. By integrating all of this knowledge, we believe it can create a foundation for innovation.”
 — Donggeun Yoo, Chief AI Officer at Lunit

Agentic Systems as the Next Phase of Medical AI

Beyond the foundation model itself, the initiative also focuses on developing an agentic system — one that can reason, plan, and learn through interaction.

“General-purpose models often struggle with highly specialized questions or complex chemical formulations. For use in expert domains, purpose-built models are essential.”
 — Donggeun Yoo, Chief AI Officer at Lunit

Within this framework, two primary use cases are envisioned:

• Healthcare: A clinical decision support system (CDSS) that responds to physicians’ questions by citing evidence from medical literature, clinical guidelines, drug data, and real-world clinical information.
• Biopharma: A conversational co-scientist that supports hypothesis generation and validation in pharmaceutical and biotechnology research through interactive dialogue.

Together, these use cases are intended to form a continuous loop connecting clinical decision-making, clinical research, and drug development, with the longer-term aim of improving patient outcomes.

“We expect this model to serve as a co-scientist in drug development and as a supporting assistant that helps physicians in clinical decision-making.”
 — Donggeun Yoo, Chief AI Officer at Lunit

Why This Required B200 GPUs at Scale

Another defining feature of the initiative is the scale of the GPU infrastructure involved.

Out of a total project budget of approximately KRW 182 billion, around KRW 174 billion has been allocated to GPU infrastructure, including 256 NVIDIA B200 (Blackwell) GPUs.

In large-scale training settings, B200 GPUs deliver significantly higher performance than A100 GPUs, and show clear performance improvements compared to H100 or H200 GPUs. At the system level, this corresponds to computing capacity comparable to hundreds of H100 GPUs or over a thousand A100 GPUs.

Deploying GPU infrastructure on this scale within a single, publicly funded medical AI initiative remains uncommon even at a global level.

Why Lunit: From SCOPE to Foundation Models

This national initiative brings together 23 organizations across industry, academia, research institutes, and hospitals. At its center, Lunit serves as the lead organization and overall technical and clinical validation coordinator.

Lunit’s work in medical AI began with early cancer detection in medical imaging and later expanded into pathology-based AI through Lunit SCOPE, which has been used to support precision oncology through treatment response prediction and quantitative biomarker analysis.

Through this work, Lunit has built several core capabilities:

• Structuring complex pathology data to support clinician review and judgment
• Operating large-scale real-world clinical data
• Integrating pathology, imaging, and clinical information
• Deploying AI systems validated in real medical environments

Seen in this context, the national project represents a natural progression rather than a departure, extending the capabilities developed through SCOPE into a full-cycle medical science foundation model.

Beyond a Single Project: Building an Ecosystem

The initiative is designed with a long-term ecosystem impact in mind.

Key elements include open-source release of core models to support adoption across Korea’s medical and biotechnology communities, alongside collaboration among consortium members on commercial-grade products and services built on top of the foundation model.

The goal is not a single technological breakthrough, but the establishment of shared infrastructure that can support sustained innovation across the medical science AI ecosystem.

Looking Ahead

This national initiative is not focused on building a single AI product. Instead, its objective is to establish a medical science foundation model and agentic system that connects research, clinical practice, and patient care into a continuous framework.

Through this effort, Lunit continues to contribute to technological progress and societal impact by building AI infrastructure that supports medical science and advances the practical application of AI in cancer care.

Glossary

• Foundation Model
  A large-scale AI model trained on broad, multimodal datasets and designed to serve as a base for multiple specialized medical and scientific applications.
• Agentic System
  An AI system capable of autonomously setting goals, planning actions, and learning through interaction.
• Clinical Decision Support System (CDSS)
  Software that assists clinicians by analyzing medical knowledge and patient-specific data to support clinical decision-making.

What an NBC Case Reveals About Today’s Breast Cancer Diagnostics

A recent NBC News report featured a powerful breast cancer screening case — AI detected a subtle abnormality that was initially invisible to the radiologist, ultimately leading to the discovery of early-stage cancer.

This case illustrates more than the value of a particular technology.
It reflects a broader shift in how breast imaging is being supported, strengthened, and modernized by AI.

A story that signaled a bigger shift

Last summer, Deirdre Hall received a routine mammogram.
Her radiologist saw no shadows, masses, or irregularities that might indicate cancer.

But AI software flagged a small region in the upper part of her left breast.
That single alert prompted further imaging and a biopsy — revealing four early-stage tumors hidden within dense, overlapping tissue.

As her radiologist noted:

“It camouflaged the cancer. Even I could have missed it.”

This wasn’t the case of AI outperforming a specialist.
It was AI compensating for structural limits of human-only interpretation.

Why dense breasts matter

About 40% of women in the United States have dense breast tissue,
which makes mammograms significantly harder to interpret.

Dense tissue can obscure early cancers, even for highly experienced radiologists.
AI helps by analyzing the entire image consistently and detecting subtle patterns that may not be visible to human readers.

Hospitals are redesigning screening

Major U.S. medical centers are integrating AI into routine clinical workflows, including:

• University of California, San Francisco (UCSF)
• MD Anderson Cancer Center
• Mount Sinai
• Perelman Center for Advanced Medicine
• MedStar Health

At UCSF, researchers found that AI-supported triage reduced the average time from mammogram to biopsy from 73 days to 9 — an 87% decrease.
For breast cancer, where delays can affect outcomes, this acceleration is especially meaningful.

Evidence that builds trust

A growing body of research demonstrates the strong real-world performance of Lunit INSIGHT DBT across diverse clinical settings:

• JAMA Oncology (8,800 cases): A large population-based study showing 88.6% accuracy with Lunit INSIGHT DBT
• Radiology: Independent reports describing cases where Lunit INSIGHT DBT helped identify findings that were not visible during the initial human read, effectively strengthening diagnostic confidence
• Consistent performance across both MMG and DBT, underscoring the robustness and reliability of the Lunit INSIGHT DBT model

Together, these findings illustrate how Lunit INSIGHT DBT enhances diagnostic consistency, reduces reader variability, and serves as a dependable second layer of analysis — supporting the expertise of radiologists.

From pilot to infrastructure

AI adoption is rapidly expanding from small-scale pilots to national screening systems.

Lunit’s AI is now deployed in:

• 65+ countries
• 10,000+ clinical institutions worldwide
• More than 200 U.S. hospitals and screening centers
• 11 public health screening regions in Italy

This scale reflects increasing confidence in AI as part of the screening infrastructure, Not merely a supplemental tool.

The role of Lunit INSIGHT DBT in this evolving landscape

In the NBC case, the AI system used was Lunit INSIGHT DBT.
DBT (Digital Breast Tomosynthesis) is particularly challenging to read in dense breast populations due to its complex 3D structure.

Lunit INSIGHT DBT consistently and precisely highlights subtle findings that are easily missed by the human eye alone.

The system does not replace radiologists;
it strengthens their ability to detect early cancers that may otherwise remain hidden.

As screening workflows evolve, Lunit continues to advance this transition in a way that is smarter, faster, and ultimately more collaborative for clinicians and patients alike.

Validated in real-world practice

Lunit INSIGHT DBT has been extensively validated through large-scale studies and real-world clinical use:

• Strong early-cancer detection performance
• Proven benefits in dense breast environments
• Robust results across equipment types and clinical settings
• Numerous cases where cancers missed during initial reads were identified by AI

These findings demonstrate that Lunit’s AI functions as a clinically reliable, operationally scalable diagnostic partner.

Why this shift matters

AI addresses long-standing challenges in breast cancer screening:

• Reducing missed cancers
• Shortening diagnostic delays
• Improving consistency across clinicians and institutions
• Enhancing detection in dense breast populations

Strengthening these areas contributes to a more reliable, equitable screening ecosystem.

A trend that’s just beginning

AI is not replacing radiologists.
But it is becoming an essential layer of diagnostic support — 
one that improves accuracy, accelerates workflows and minimizes the structural blind spots of traditional screening.

The NBC case is one example of this growing shift.
As adoption accelerates, AI will become a standard component of early detection workflows.

Lunit will continue advancing this transition through proven, real-world-ready AI solutions and through this new blog, we hope to share more real-world insights that show how AI is reshaping cancer detection worldwide.

Source: NBC News — Artificial intelligence flagged woman’s mammogram, helping to catch her breast cancer early
https://www.nbcnews.com/health/cancer/artificial-intelligence-mammogram-scan-breast-cancer-results-reading-rcna237293