Shallow and deep learning classifiers in medical image analysis.

Area of Science:

• Medical Artificial Intelligence
• Machine Learning in Healthcare
• Radiology Informatics

Background:

• The synergy between artificial intelligence (AI) and medicine is rapidly advancing predictive modeling for physician decision-making.
• Machine learning (ML), a sub-branch of AI, is pivotal in developing these clinical decision support systems.
• Understanding ML classifiers is essential for their application in healthcare, particularly in radiology.

Purpose of the Study:

• To provide educational insights into accessible and widely used machine learning classifiers in radiology.
• To differentiate between traditional (shallow) and deep learning architectures.
• To offer guidelines for selecting appropriate classifiers based on specific clinical needs and resources.

Main Methods:

• Review and comparison of shallow learning algorithms (Support Vector Machines, Random Forest, XGBoost).
• Review and comparison of deep learning architectures (Convolutional Neural Networks, Vision Transformers).
• Discussion of key steps in classifier training and feature extraction (e.g., radiomics for shallow learning).

Main Results:

• Shallow classifiers require manual feature extraction from regions of interest.
• Deep classifiers automate feature extraction and classification processes.
• Classifier selection depends on task requirements, dataset size, explainability needs, and computational resources.

Conclusions:

• AI and ML, particularly various classifiers, offer significant potential for enhancing clinical decision support in radiology.
• The choice of classifier involves balancing performance with practical considerations like data availability and interpretability.
• Explainability of AI models is crucial for their trustworthy integration into clinical practice.