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

Abstract

Raman Spectroscopy has long been anticipated to augment clinical decision making, such as classifying oncological samples. Unfortunately, the complexity of Raman data has thus far inhibited their routine use in clinical settings. Traditional machine learning models have been used to help exploit this information, but recent advances in deep learning have the potential to improve the field. However, there are a number of potential pitfalls with both traditional and deep learning models. We conduct a literature review to ascertain the recent machine learning methods used to classify cancers using Raman spectral data. We find that while deep learning models are popular, and ostensibly outperform traditional learning models, there are many methodological considerations which may be leading to an over-estimation of performance; primarily, small sample sizes which compound sub-optimal choices regarding sampling and validation strategies. Amongst several recommendations is a call to collate large benchmark Raman datasets, similar to those that have helped transform digital pathology, which researchers can use to develop and refine deep learning models.

Discovering Glioma Tissue through Its Biomarkers’ Detection in Blood by Raman Spectroscopy and Machine Learning

“…A Python library BaselineRemoval contains several methods to remove the noise, caused by fluorescence in Raman spectra, and according to our previous findings [ 54 ], the airPLS algorithm [ 51 ] has a good performance. Principal component analysis (PCA) was used to extract informative features [ 55 , 56 ].…”

Deep Learning Applied to Raman Spectroscopy for the Detection of Microsatellite Instability/MMR Deficient Colorectal Cancer

“…This allows for hyperparameters to be optimised in an inner CV loop, and then tested against held-out data in an outer CV loop ( Figure 1 ). Nested CV has been shown to reduce a model’s estimated accuracy by as much as 20% in oncological applications of RS, giving a more realistic assessment of the model’s generalisability [ 8 ]. A single metric needs to be selected to optimise: the log loss was chosen as it is a proper scoring metric which utilises distributional information, compared to typical binarised metrics such as accuracy [ 17 ].…”

“…A Raman spectrum represents the change in photon wavenumber from a monochromatic light source along the x-axis and the intensity (i.e., number of photons) thus scattered on the y-axis. RS has been successfully applied to discriminate between numerous cancer types in human tissues, most recently to the brain, breast, cervical, colon, lung, nasopharyngeal, prostate, skin and tongue [ 8 ]. The applications include early diagnosis, biopsy guidance and intraoperative tumour margin detection.…”

“…Its ability to capture non-linear relationships make it suited to complex clinical datasets, and it may help unleash the potential of RS. In addition to being increasingly applied to oncology problems in the context of digital histopathology [ 10 ], it is now extending into oncological applications of RS [ 8 ]. However, DL is notoriously data-intensive, which makes it difficult to apply to smaller medical proof-of-concept studies.…”

Open-sourced Raman spectroscopy data processing package implementing a baseline removal algorithm validated from multiple datasets acquired in human tissue and biofluids

“…A strength of the technique is that this information can be reinterpreted as a molecular fingerprint lending an interpretation of the material’s composition in terms of the relative concentration of proteins and specific amino acids, lipids, deoxyribonucleic acid, and ribonucleic acid, as well as water and other metabolites. 4 , 5 …”