Machine learning techniques for biomedical image segmentation: An overview of technical aspects and introduction to...

Area of Science:

• Computer Science, Artificial Intelligence
• Radiology, Nuclear Medicine & Medical Imaging

Background:

• Machine learning (ML) algorithms are crucial for advancing medical image segmentation.
• Accurate and efficient segmentation is vital for medical diagnosis and treatment planning.

Purpose of the Study:

• To review the role of ML algorithms in medical image segmentation.
• To highlight key studies applying ML to biomedical image segmentation.
• To discuss classical and deep learning methods, their successes, limitations, and training challenges.

Main Methods:

• Review of classical machine learning algorithms (e.g., Markov random fields, k-means, random forest).
• Review of deep learning architectures (e.g., artificial neural networks, convolutional neural networks, recurrent neural networks).
• Analysis of segmentation results from studies published in the last three years.

Main Results:

• Classical ML models offer sample efficiency and simpler structures, though often less accurate than deep learning.
• Deep learning techniques demonstrate high accuracy in medical image segmentation.
• Successes and limitations of various ML paradigms in biomedical image segmentation are identified.

Conclusions:

• Machine learning, encompassing both classical and deep learning approaches, is essential for accurate and efficient medical image segmentation.
• Understanding the trade-offs between different ML models is key for optimal application.
• Addressing training challenges through heuristics can further enhance ML model performance in medical imaging.