On 2013 Nov 04, Allison Stelling commented:

This was a good read. The authors do a good job going over the present body of literature about using Raman and IR for medical diagnostics. I particularly liked the fact that they defined "sensitivity" and "specificity" in their Introduction, and kept the statistics and processing algorithms reasonably straightforward in their analysis. I also admire the fact that they used two different spectral methods (florescence and Raman spectroscopy) to investigate the tissue; combining the two increases the likelihood of accurate and reliable diagnostic results.

I wonder, though, if it would be more effective to individualize the diagnosis even further- that is, take a background spectra on a "healthy" section, and take a huge number of acquisitions. The disease phenotype might pop out in the difference spectra as the instrument scans over the tissue area. (More likely, many things will. But, you might be able to train it on the disease's signature, and detect it within the subtraction spectra.) This could possibly avert the issue of training sets with misclassified (misdiagnosed) spectra in the training set for "healthy" (or vice versa). With tumors, there can be quite of lot of inter-individual variance within different tumor classifications and subtypes. Raman could really help with more tailored, individualized diagnostics.

On 2013 Nov 04, Allison Stelling commented:

This was a good read. The authors do a good job going over the present body of literature about using Raman and IR for medical diagnostics. I particularly liked the fact that they defined "sensitivity" and "specificity" in their Introduction, and kept the statistics and processing algorithms reasonably straightforward in their analysis. I also admire the fact that they used two different spectral methods (florescence and Raman spectroscopy) to investigate the tissue; combining the two increases the likelihood of accurate and reliable diagnostic results.

I wonder, though, if it would be more effective to individualize the diagnosis even further- that is, take a background spectra on a "healthy" section, and take a huge number of acquisitions. The disease phenotype might pop out in the difference spectra as the instrument scans over the tissue area. (More likely, many things will. But, you might be able to train it on the disease's signature, and detect it within the subtraction spectra.) This could possibly avert the issue of training sets with misclassified (misdiagnosed) spectra in the training set for "healthy" (or vice versa). With tumors, there can be quite of lot of inter-individual variance within different tumor classifications and subtypes. Raman could really help with more tailored, individualized diagnostics.