Introduction to supervised machine learning in clinical epidemiology.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Computational Biology

Background:

• Machine learning (ML) has advanced significantly due to increased computing power and health data availability.
• ML encompasses supervised, unsupervised, and reinforcement learning, each with distinct medical applications.
• Supervised learning is prevalent for medical diagnoses and prognoses.

Purpose of the Study:

• To explain the concepts and applications of supervised learning in medicine.
• To provide an overview of four key predictive algorithms: random forests, gradient-boosted decision trees, support vector machines, and neural networks.
• To highlight the role of deep learning, an extension of neural networks, in complex medical tasks like image analysis.

Main Methods:

• Focus on supervised learning algorithms.
• Explanation of random forests, gradient-boosted decision trees, support vector machines, and neural networks.
• Discussion of deep learning for image processing, exemplified by diabetic retinopathy diagnosis from retinal fundus photographs.

Main Results:

• Supervised learning is widely applied for medical diagnoses and prognoses.
• Neural networks have evolved into deep learning, capable of handling complex tasks like medical image analysis.
• Algorithms like random forests, gradient-boosted decision trees, and support vector machines are effective for prediction.

Conclusions:

• Machine learning provides novel insights by processing large datasets beyond human capacity.
• Neglecting domain knowledge can lead to algorithm failure in medical applications.
• Successful implementation of ML in medicine requires integrating algorithms with human cognitive abilities.