Figure 6
From: Non-invasive screening of breast cancer from fingertip smears—a proof of concept study
Dimensionality reduction of spectra to 2D using (i-iii) PCA and (iv-vi) UMAP algorithms. Results are illustrated for three independent spectra scaling approaches, applied prior to dimensionality reduction: (i, iv) no additional scaling, (ii, v) standardisation of each m/z peak position to mean = 0, standard deviation = 1 across the n = 135 dataset, and (iii, vi) min–max scaling where each m/z peak position is scaled to a fixed [0,1] domain. In each subplot, each scatter point corresponds to an individual sample in the n = 135 dataset, coloured by cancer diagnosis state. The three benign samples that were misclassified by the MLP model as “metastatic” and “early cancer” have been highlighted as cross and square markers, respectively.
