Diagnosing Breast Cancer with A.I. | Meyyammai Meyyappan & Sabari Laxmi | TEDxPOWIIS Youth

Key Concepts:

• Breast cancer diagnosis
• Machine learning in medical diagnosis
• Cell morphology analysis
• Fine needle aspiration (FNA)
• Benign vs. malignant tumors
• Wisconsin breast cancer data set
• Correlation vs. causation in AI
• Logistic regression
• Support vector machine (SVM)
• Random forest
• Gradient boosting
• AI in medicine

1. Introduction: The Urgency of Breast Cancer Diagnosis

• Every 14 seconds, a woman is diagnosed with breast cancer, highlighting the critical need for timely and accurate diagnosis.
• The video emphasizes the potential of AI to improve diagnostic speed and accuracy, potentially saving lives.
• The case of "Patient X," a 44-year-old woman with two tumors (one malignant, one benign), is introduced to illustrate the complexities of diagnosis and the potential impact of AI.

2. The Limitations of Traditional Diagnostic Methods

• Traditional breast cancer diagnostics can be lengthy, expensive, and subjective.
• Pathologist interpretations, while highly trained, can vary due to fatigue, variability, and potential for error.
• The project aims to use machine learning to analyze cell morphology and predict whether cells are benign or malignant.

3. Cell Morphology and Cancer Detection

• Healthy cells are generally regular, while cancerous cells exhibit irregularities:
  • Irregular shape
  • Elongated structure
  • Swollen nucleus (due to excess DNA activity)
• Fine Needle Aspiration (FNA): A common, relatively painless procedure to collect cell samples for analysis.
• The machine learning model provides specific measurements (radius, area, texture, perimeter) instead of subjective assessments.

4. Key Features of the Machine Learning Model: Cell Morphology

• The model focuses on cell morphology, specifically:
  • Radius: Distance from the nucleus center to the edge (uneven in cancer cells).
  • Perimeter: Total boundary of the nucleus (irregular in cancer cells).
  • Texture: Smoothness or graininess of the nucleus (uneven and grainy in cancer cells).
  • Area: Cancer cells have a larger area than normal cells.

5. The Wisconsin Breast Cancer Data Set

• The model was trained using the Wisconsin breast cancer data set, a reliable benchmark for medical algorithm training.
• The data set contains 569 samples with 30 key features and 10 cell attributes.
• Created by Dr. William Walberg in the 1980s, the data set is based on FNA samples from breast cancer patients, with confirmed benign or malignant diagnoses.
• This provides a supervised learning data set for effective model training.

6. Why AI is Suitable for Cancer Diagnosis: Correlation vs. Causation

• AI excels at identifying correlations between data points, which is crucial when time is limited.
• The video explains the difference between correlation and causation using the example of ice cream sales and drowning accidents on a hot day.
• In the context of cancer, AI can correlate multiple variables (e.g., nuclear perimeter, fractal dimension) to predict malignancy.

7. Model Implementation: Statistical Concepts

• The model uses a combination of statistical concepts:
  • Logistic Regression: Suitable for variables with strictly positive or negative correlations (e.g., nuclear perimeter).
  • Support Vector Machine (SVM): Categorizes variables and uses distance from a hyperplane to map predictions.
  • Random Forest and Gradient Boosting: Use decision trees and probabilities to determine the most likely outcome.
• The model is trained on 80% of the data set, creating a map of data points for comparison.

8. Patient X's Results and the Importance of AI

• The model correctly identified the malignancy probability of Patient X's tumors:
  • Right breast tumor (malignant): >98% probability.
  • Left breast cyst (benign): ~2% probability.
• Due to the lack of AI accessibility, Patient X underwent a mastectomy and passed away within 4 months.
• The video questions the extent to which humans can perform before AI intervention could improve the process.

9. The Future of AI in Medicine

• AI is the future of medicine, and its capabilities and limitations must be understood.
• The vision is a future where nurses in rural areas can use AI to get a second opinion almost instantly.
• The project aims to assist doctors and pathologists, not replace them, by providing a second pair of eyes trained on vast amounts of data.

10. Conclusion: Saving Lives Through AI

• The main takeaway is that AI has the potential to significantly improve breast cancer diagnosis by increasing speed, accuracy, and accessibility.
• By leveraging machine learning and cell morphology analysis, the project aims to save lives by catching what might be missed by traditional methods.