The authors conducted a comprehensive evaluation of various machine learning models for classifying glioblastoma cells using Raman spectroscopy data. We compared traditional machine learning methods such as support vector machines, boosting, and random forests with more modern methods such as convolutional neural networks, visual transformers, and generalized learning systems. The study found that convolutional neural networks were the most suitable models for this dataset, performing best without background drift removal or other preprocessing methods, but with some feature normalization. Additionally, this work analyzed and compared the impact of baseline data processing on model classification performance. Data augmentation was applied to the data, and the same machine learning models were used for evaluation. The experiments revealed that data augmentation did not improve model performance, contrary to findings in other published works in this field. This study explored some specific challenges in applying Raman spectroscopy to machine learning, which holds positive implications for the integration of Raman spectroscopy and machine learning in medical applications. The article has some issues, but it could be considered for publication after major revisions.
1. In the first part of the introduction, the author mentions that Raman spectroscopy has become a very promising tool for cancer diagnosis. However, the references cited by the author are somewhat outdated, and it is recommended to update them. Specific references can be found in the following literature:
(1) DOI:10.1016/j.aca.2024.343302
(2) DOI:10.1016/j.artmed.2024.103053
(3) DOI:10.1016/j.chemolab.2023.104762
2. Have the authors considered the significantly increased complexity of Raman signals in more complex systems (e.g: bodily fluids)? In such cases, does the results of this study have universal applicability?
3. The authors collected Raman spectra from three cell types: cancer cells, monocytes, and T-cells (Figure 1). The authors mentioned the relationship between gene expression in these three cell types within the body. Can the authors explain this relationship based on the characteristic peaks in the Raman spectra?
4. Although the authors used machine learning models to accurately classify the spectral data, it is still necessary to assign the characteristic peaks of the spectra.
5. The author's reference format has significant issues.
6. The author mentions in the introduction and experimental description that two experiments were designed in the article, but the description of the two experiments in the writing and description of the article is too vague, which may cause great confusion to readers.
7. The results section of this article is more like an experimental report and does not effectively analyze and describe the results of the experiment. We hope that this section will be improved.
8. The illustrations in the article are not aesthetically pleasing and should be improved.