Statistics of Generative Artificial Intelligence and Nongenerative Predictive Analytics Machine Learning in Medicine.

Area of Science:

• Medicine
• Artificial Intelligence
• Machine Learning
• Statistics

Background:

• The integration of artificial intelligence (AI) and machine learning (ML) in medicine necessitates a strong understanding of their statistical underpinnings.
• Pathology and medicine generate extensive data suitable for AI/ML applications, with generative AI, particularly large language models, emerging as transformative tools.

Purpose of the Study:

• To provide an overview and comparative analysis of statistical measures used in both generative AI and traditional (non-generative predictive analytics) ML.
• To highlight the strengths and limitations of these statistical methodologies for medical applications.

Main Methods:

• Review of statistical measures commonly employed in generative AI (e.g., perplexity, BiLingual Evaluation Understudy score).
• Review of statistical measures used in traditional ML for classification (e.g., accuracy, sensitivity, F1 score, AUC) and regression (e.g., RMSE, R²).
• Comparative analysis of the identified statistical measures, focusing on their applicability and limitations in medical contexts.

Main Results:

• Generative AI utilizes unique metrics like perplexity and BiLingual Evaluation Understudy score for sample quality assessment, which are often unfamiliar to medical practitioners.
• Traditional ML employs more familiar metrics such as accuracy, sensitivity, F1 score, AUC for classification, and RMSE, R² for regression.

Conclusions:

• Understanding the similarities and differences in statistical measures between generative and traditional ML is crucial for medical professionals.
• Informed application of these statistical principles will enable responsible and scientifically sound advancements in AI/ML within medicine, addressing current and future challenges.