Case Study 1: Early Lung Cancer Detection

The Medical Challenge

Lung cancer is the leading cause of cancer death worldwide, but early detection dramatically improves survival rates. When caught at Stage I (very early), the 5-year survival rate is approximately 60%. However, if not detected until Stage IV (advanced), the 5-year survival rate drops to just 6%.

The Problem:

Small lung nodules (tiny masses) visible on CT scans can be incredibly difficult for radiologists to interpret. Many small nodules are benign (not cancerous), but some are early-stage cancers. Radiologists must examine hundreds of images per scan, looking for nodules that may be only a few millimeters in size. Studies show that even experienced radiologists miss small lung nodules in 20-30% of cases.

Maria's primary care doctor ordered a chest CT scan to investigate her persistent symptoms. The scan generated 300+ cross-sectional images of her lungs that needed careful examination.

The AI System in Action

Maria's hospital uses an FDA-approved AI system called DeepRadiology Lung Scanner, which was developed specifically to detect lung nodules and assess cancer risk.

How the AI System Works:

Training Phase (Before Maria's Scan): The AI was trained on over 45,000 chest CT scans from patients with confirmed diagnoses. For each scan, radiologists had marked all nodules and indicated whether they were later confirmed as cancerous or benign. The machine learning algorithm learned to recognize visual patterns associated with malignant (cancerous) versus benign nodules.
Analysis Phase (Maria's Scan): The AI system analyzed all 300+ images from Maria's CT scan in approximately 8 seconds. It used computer vision algorithms to detect any abnormal masses and deep learning neural networks to assess the likelihood that each detected nodule was cancerous.
Risk Assessment: For each nodule detected, the AI calculated a cancer probability score from 0-100%. It also compared the nodule's size, shape, location, and texture patterns against thousands of similar cases in its training data.

The AI's Findings:

The system detected a 7mm nodule in Maria's right upper lung lobe that it flagged with a 78% cancer probability score. This small nodule had several characteristics associated with early-stage adenocarcinoma (a type of lung cancer):

Irregular, spiculated (spiky) edges rather than smooth borders
Heterogeneous (varied) internal density
Location in the peripheral lung tissue (common for adenocarcinoma)
Slight enlargement compared to a previous scan from 2 years earlier

Critical Detail: The initial human radiologist who reviewed Maria's scan had noted the nodule but classified it as "likely benign" and recommended routine follow-up in 6 months. The AI's high-risk flag prompted a second radiologist to perform a more detailed review, who agreed with the AI's assessment and recommended immediate biopsy.

Outcome and Treatment

A biopsy confirmed early-stage lung cancer (Stage IA). Because it was caught so early:

Maria underwent minimally invasive video-assisted thoracoscopic surgery (VATS) to remove the tumor
The cancer had not spread to lymph nodes or other organs
No chemotherapy or radiation was required
Maria recovered fully within 6 weeks
Her 5-year survival probability is approximately 70%

Comparison to Alternative Scenario: If the nodule had gone undetected for 6 more months (following the initial radiologist's recommendation), it likely would have grown larger and possibly spread. At that point, Maria would have required more aggressive surgery, chemotherapy, and/or radiation, with significantly lower survival odds.

Measure	Human Radiologist	AI System	Human + AI Together
Detection Rate (Sensitivity)	91%	94%	96%
False Positive Rate	6.5%	5.8%	4.2%
Time per Scan	10-15 minutes	8 seconds	10-15 minutes
Consistency	Varies (fatigue, time of day)	Consistent	High

Important Considerations

What the AI Did Well:

Identified a subtle nodule that might have been overlooked
Accurately assessed high cancer risk based on visual patterns
Provided objective, quantitative risk score
Worked quickly and consistently without fatigue
Prompted more thorough human review

Limitations and Important Points:

The AI was a decision support tool, not the final decision-maker. Radiologists reviewed the AI's findings and made the ultimate recommendations.
The AI does not replace the need for skilled radiologists—it augments their capabilities
The system was trained on diverse patient populations but may still have biases based on training data
False positives (flagging benign nodules as high-risk) can cause patient anxiety and lead to unnecessary biopsies
The AI cannot consider patient preferences, overall health status, or psychological factors—only image patterns
Not all hospitals have access to this expensive technology

Case Study Analysis Worksheet

Student Name: ___________________ Date: _______________

Question 1: Understanding the Medical Problem

Why is early detection of lung cancer so critical? Use specific statistics from the case study to support your answer.

Question 2: AI System Mechanics

Explain in your own words how the AI system learned to identify lung cancer. What data did it need during training, and what did it learn from that data?

Question 3: Comparing Performance

According to the data table, what are two advantages of using AI alongside human radiologists compared to using either alone?

Question 4: Critical Thinking About AI Role

The case study mentions that the AI flagged a nodule the human radiologist initially thought was "likely benign." However, AI systems sometimes flag benign nodules as high-risk (false positives). Discuss the trade-offs: Is it better for an AI system to be overly cautious (more false positives) or miss some cancers (more false negatives)? Explain your reasoning.

Question 5: Human-AI Collaboration

Why didn't the hospital simply let the AI make the diagnosis without human radiologist review? What can human doctors do that AI cannot?

Question 6: Accessibility and Equity

The case study notes that "not all hospitals have access to this expensive technology." What are the potential consequences of some hospitals having AI diagnostic tools while others do not? How might this affect healthcare equality?

Question 7: Patient Perspective

If you were Maria, how would you feel knowing that an AI system identified your cancer? Would it increase or decrease your confidence in the diagnosis? Why?

Question 8: Future Implications

Based on this case study, predict one way that AI might further improve lung cancer detection in the next 10 years. Be specific and explain why your prediction would be beneficial.

Discussion Questions for Groups

Should health insurance companies be required to cover AI-assisted diagnostic tools if they improve detection rates? Why or why not?
If an AI system and a human radiologist disagree about a diagnosis, who should have the final say?
What information would you want to know about an AI system before trusting it to help diagnose your medical condition?
How might AI lung cancer detection affect healthcare costs? Consider both the cost of the technology and the cost savings from earlier detection.

Vocabulary Terms

Benign: Not cancerous; not harmful
Malignant: Cancerous; capable of spreading and causing harm
Nodule: A small mass or lump of tissue
Computer Vision: AI technology that enables computers to interpret and understand visual information from images
Deep Learning Neural Networks: Machine learning systems modeled on the human brain, capable of learning complex patterns from large datasets
Sensitivity (True Positive Rate): The ability of a test to correctly identify disease when it is present
False Positive: When a test incorrectly indicates that a disease is present
False Negative: When a test incorrectly indicates that a disease is not present
Adenocarcinoma: A type of cancer that forms in mucus-secreting glands; the most common type of lung cancer

Medical Case Study 1: Early Lung Cancer Detection with AI

Patient Background