Machine Learning of Raman Spectroscopy Data for Classifying Cancers: A Review of the Recent Literature.

Area of Science:

• Biomedical Optics
• Computational Pathology
• Spectroscopic Analysis

Background:

• Raman spectroscopy offers potential for clinical decision-making, particularly in oncological sample classification.
• The inherent complexity of Raman data has limited its routine clinical adoption.
• Machine learning, including deep learning, is being explored to harness Raman spectral information.

Purpose of the Study:

• To review recent machine learning methods for cancer classification using Raman spectral data.
• To identify potential pitfalls in current deep learning and traditional machine learning approaches.
• To provide recommendations for improving the reliability of these methods in clinical settings.

Main Methods:

• A comprehensive literature review was conducted on machine learning applications in Raman spectroscopy for cancer classification.
• Analysis focused on identifying common methodologies, performance metrics, and potential biases.
• The review examined both traditional machine learning and deep learning models.

Main Results:

• Deep learning models are increasingly prevalent and often reported to outperform traditional models.
• Methodological concerns, particularly small sample sizes, may lead to overestimated performance.
• Suboptimal sampling and validation strategies were identified as significant issues compounding small sample sizes.

Conclusions:

• While deep learning shows promise, current performance estimates may be inflated due to methodological limitations.
• There is a critical need for larger, standardized benchmark Raman datasets.
• Developing robust deep learning models requires addressing issues of sample size, sampling, and validation strategies for reliable clinical translation.