Traditional machine learning in biomedical image analysis: before you go too deep

Area of Science:

• Biomedical image analysis
• Machine learning
• Medical imaging

Background:

• Traditional machine learning (TML) algorithms offer advantages in multimodal data integration, interpretability, computational efficiency, and robustness on smaller datasets.
• Deep learning (DL) currently dominates medical image analysis due to superior performance and end-to-end feature learning.
• Despite DL's prevalence, TML retains unique value in specific biomedical imaging applications.

Purpose of the Study:

• To provide a comprehensive review of TML applications across various biomedical imaging modalities.
• To highlight the core principles, practical implementation, and benefits of TML in the context of deep learning.
• To analyze the continued relevance and unique value proposition of TML in medical image analysis.

Main Methods:

• Review of fundamental machine learning concepts relevant to biomedical imaging.
• Description of key biomedical imaging tasks addressed by TML.
• Analysis of prominent examples showcasing TML's strengths.
• Identification of popular platforms enabling TML utilization by clinicians and researchers.

Main Results:

• TML is effective for multimodal data processing, applications with limited data, and scenarios requiring high interpretability.
• TML offers advantages in computational efficiency and robustness, particularly on smaller datasets.
• Democratized tools and validated clinical studies support the continued use of TML.

Conclusions:

• Traditional machine learning remains a vital methodology for extracting quantitative and qualitative insights from biomedical image data.
• TML offers unique value in specific niches of medical image analysis, complementing deep learning approaches.
• The continued relevance of TML in both research and clinical practice is ensured by its inherent strengths and increasing accessibility.