# AI in Lung Cancer Screening
Source: https://humanspark.ai/ai-in-lung-cancer-screening/

## **A Breakthrough in Reducing Radiologist Workload**

Artificial intelligence (AI) is already transforming healthcare, and a recent study published in the **European Journal of Cancer** confirms its effectiveness in lung cancer screening.

Researchers validated a commercially available AI system using data from the **UK Lung Cancer Screening (UKLS) trial**, showing that AI can:

✅ **Reduce radiologist workload by up to 79%** by ruling out low-risk CT scans.
 ✅ Achieve a **99.8% negative predictive value (NPV)** - ensuring almost no cancers are missed.
 ✅ **Match or exceed human radiologists in accuracy**, with fewer misclassifications.
 ✅ **Detect all 31 baseline-round lung cancers**, though one was initially classified as negative due to a size threshold.
 ✅ **Improve sensitivity by adjusting the AI's volume threshold** - lowering it from **100mm³ to 80mm³** would have eliminated the single misclassification.

The study's authors conclude that AI has the potential to significantly reduce workload without compromising diagnostic accuracy, making it a valuable tool for large-scale lung cancer screening programs.

## **The Urgent Need for AI in Lung Cancer Screening**

Lung cancer remains the **leading cause of cancer-related deaths worldwide**, with survival rates directly tied to early detection. **Low-dose CT (LDCT) screening** is one of the most effective ways to catch lung cancer early, but implementing large-scale screening programs requires significant radiologist resources.

🚨 **The Problem? A global shortage of radiologists is limiting screening availability.**

According to a report by the Royal College of Radiologists, the UK alone has a **33% shortfall of radiologists**, making it difficult to scale lung cancer screening to the levels seen in the **U.S. National Lung Screening Trial (NLST)** or **NELSON trial** in Europe. AI offers a promising solution by **automating the review of low-risk scans**, allowing radiologists to focus on cases that require expert attention.

The study found that AI can serve as an efficient first-reader, reducing the number of scans requiring human review by up to 79%, helping to ease the burden on radiologists while maintaining high diagnostic accuracy.

## **How the Study Validated AI Performance**

This study assessed an AI system using a **sequestered dataset** from the **UKLS trial**, meaning the AI had never encountered these scans before - eliminating the risk of "overfitting". The researchers compared the AI's performance against:

✔️ **Expert panel reference standards** (considered the gold standard for accuracy)
 ✔️ **Human radiologists of varying experience levels**

### **Key Findings:**

🔹 **AI ruled out up to 79% of scans**, reducing the radiologist workload without compromising accuracy.
 🔹 **AI outperformed human radiologists**, showing fewer misclassifications when benchmarked against the expert panel.
 🔹 **All 31 baseline-round lung cancers were detected**, though one was initially misclassified due to a **100mm³ volume threshold**.
 🔹 **Adjusting the threshold to 80mm³ would have eliminated this misclassification**, ensuring **99.8% NPV**.

The study highlights that maintaining a high negative predictive value is crucial for screening programs, and with an NPV of 99.8%, AI presents a clinically viable solution that could help scale lung cancer screening worldwide.

## **How AI Works in Lung Cancer Screening**

The AI system doesn't replace radiologists completely - it acts as a **first-reader**, ruling out scans that show **no nodules or only very small nodules (<100mm³)**. The process works as follows:

1️⃣ **AI analyses all CT scans**
 2️⃣ **Scans with no suspicious nodules are flagged as negative**
 3️⃣ **Radiologists focus only on the remaining scans** that need expert review

By filtering out clear cases, AI allows radiologists to **dedicate more time to complex cases**, optimizing workflow and improving patient care.

## **Challenges & Considerations for AI Implementation**

While the results are promising, implementing AI in lung cancer screening requires careful planning:

🔸 **Population Diversity** - The UKLS dataset was predominantly **white and male**, meaning further validation is needed in diverse populations.
 🔸 **Follow-Up Screening** - The study focused on **baseline screening**; AI's performance in **longitudinal follow-up scans** still needs evaluation.
 🔸 **Handling Incidental Findings** - The current AI model does not fully account for **other abnormalities** that may appear in scans.
 🔸 **Customising Thresholds** - AI's sensitivity can be adjusted by modifying volume thresholds, but **clinical protocols need to define optimal settings**.

The study emphasises that AI is not a replacement for radiologists but a tool to enhance efficiency. Successful deployment will require careful integration with existing clinical workflows and regulatory approvals.

## **What This Means for AI Adoption in Other Industries**

The success of AI in lung cancer screening highlights broader lessons for **business leaders in any industry**:

✔️ **AI excels when given a clear, defined task** - The system worked because it was focused on **ruling out normal scans**, not diagnosing disease.
 ✔️ **AI should complement, not replace, human expertise** - The best results came from **AI working alongside radiologists**, not instead of them.
 ✔️ **Validation is essential before scaling** - The AI model was tested rigorously on an **unseen dataset** before considering wider deployment.
 ✔️ **AI optimizes expert resources** - By reducing repetitive tasks, AI enables specialists to **focus on higher-value work**.

## **How This is Relevant for Businesses**

### 1️⃣ **AI Reduces Workload Without Replacing Experts**

Just as AI reduces radiologists' workload by filtering routine cases, businesses can use AI to automate repetitive, low-risk tasks. This frees up skilled employees to focus on higher-value work, improving efficiency and job satisfaction.

### 2️⃣ **AI Excels in Defined, High-Impact Roles**

The AI in this study worked well because it had a clear, well-defined task: ruling out non-cancerous scans. Business leaders should apply AI in areas where automation delivers immediate impact - like customer service chatbots, fraud detection, or supply chain optimization: - without overcomplicating its role.

### 3️⃣ **Validation Before Scaling**

The study tested AI against unseen data before considering real-world deployment. Companies should take the same approach: pilot AI tools in a controlled setting, measure success, then scale responsibly.

### 4️⃣ **AI as an Efficiency Multiplier**

The AI system didn't replace radiologists; it helped them work smarter. Businesses should see AI as an augmentation tool rather than a replacement, using it to enhance productivity rather than cut headcount.

### 5️⃣ **Industry-Specific AI Opportunities**

Just as AI streamlines lung cancer screening, similar AI-driven automation could optimise areas like:

- **Customer Insights** - AI-driven analytics tools can identify trends and customer needs faster than manual reporting.
- **Finance & Compliance** - AI can scan invoices, flag anomalies, and automate routine audits.
- **Operations & Logistics** - AI can predict demand and optimise inventory, reducing waste and improving supply chain efficiency.

### 🔍 **The Real Question**

For business leaders, the big question is: **Where can AI best reduce workload in your company while maintaining (or improving) quality?**

Start by identifying repetitive, rule-based tasks where AI can assist, then test, measure, and refine before scaling.

📄 *Read the Full Study:* *Histological Proven AI Performance in the UKLS CT Lung Cancer Screening Study: Potential for Workload Reduction* - [*Read the full paper here*](https://www.ejcancer.com/article/S0959-8049(25)00105-4/fulltext)

### **What's Your Take?**

Would you trust AI to assist in medical diagnostics? How do you see AI transforming your industry? **Let's discuss in the comments!** 👇