AI Glossary

This glossary provides key terms relevant to endocrinology, supporting informed use of emerging technologies in diagnosis, risk stratification, and disease management.

Algorithmic Bias

Systematic error arising from non-representative training data, potentially leading to disparities in endocrine care (eg, underperformance in certain populations).

Artificial Intelligence (AI)

Computational systems designed to perform cognitive tasks such as pattern recognition, classification, and prediction, increasingly applied in endocrine diagnostics and management.

Calibration

The agreement between predicted probabilities and observed outcomes, essential for reliable risk prediction models.

Clinical Decision Support Systems (CDSS)

AI-enabled platforms that provide evidence-based recommendations, such as insulin titration algorithms or thyroid nodule management pathways.

Closed-Loop Systems (Artificial Pancreas)

Automated insulin delivery systems that use real-time CGM data and control algorithms to adjust insulin dosing with minimal user input.

Computer Vision

AI methods for interpreting medical images, including thyroid ultrasound classification and automated detection of diabetic retinopathy.

Continuous Glucose Monitoring (CGM) Analytics

Advanced analytics, often ML-driven, applied to CGM data to detect glycemic variability, predict hypo-/hyperglycemia, and guide therapy optimization.

Deep Learning

A subset of ML using multilayer neural networks capable of high-dimensional feature extraction, particularly effective in imaging (eg, thyroid ultrasound interpretation, retinal screening).

Digital Biomarkers

Quantifiable physiological or behavioral data (eg, glucose variability, activity patterns) collected via digital devices and analyzed using AI for disease monitoring.

Digital Therapeutics (DTx)

Regulated, software-based interventions—often incorporating AI—for managing conditions such as diabetes, obesity, and metabolic syndrome.

Explainable AI (XAI)

Methods that make model outputs interpretable, enabling clinicians to understand and trust AI-driven recommendations.

External Validation

Assessment of model performance in independent populations, critical for clinical adoption in endocrinology.

Feature Engineering

The process of selecting and transforming variables (eg, HbA1c trends, BMI, CGM metrics) to improve model performance.

Machine Learning (ML)

A class of statistical models that learn patterns from structured and unstructured data to generate predictions, commonly used for diabetes risk modeling and treatment response prediction.

Model Training and Validation

• Training: Fitting the model on a dataset

• Validation: Testing performance on independent data to assess generalizability

Natural Language Processing (NLP)

AI techniques that extract structured data from unstructured clinical text (eg, identifying endocrine diagnoses, medication use, or complications from EHR notes).

Neural Networks

Computational architectures that model complex nonlinear relationships in data, widely used in predictive modeling for endocrine outcomes.

Overfitting

A modeling error in which an algorithm performs well on training data but poorly on new data due to excessive complexity.

Precision Endocrinology

The integration of AI with clinical, genomic, and lifestyle data to individualize diagnosis and treatment strategies.

Predictive Modeling

The application of ML to estimate future clinical outcomes, such as progression to diabetes, fracture risk, or likelihood of treatment response.

Risk Stratification Models

AI-driven tools that classify patients into risk categories for complications (eg, cardiovascular risk in diabetes, osteoporosis-related fractures).

Supervised Learning

An ML approach trained on labeled datasets (eg, known outcomes), used in endocrine applications such as predicting hypoglycemia or classifying thyroid nodules.

Time-in-Range (TIR) Optimization Algorithms

AI-based approaches that use CGM data to maximize time within target glucose ranges while minimizing hypoglycemia.

Unsupervised Learning

Algorithms that identify hidden patterns in unlabeled data, useful for phenotyping heterogeneous endocrine disorders (eg, subtypes of type 2 diabetes).