Figure 6: Classification of white blood cells (OT-II) and cancer cells (SW-480) by TS-QPI label-free features.

(a) Training process of the neural network leads to improvement of classification accuracy over generations of genetic algorithm. In addition to multivariate analysis using all 16 biophysical features extracted from the TS-QPI quantitative images (blue curves), we also show training process by three single features. Red, green, and orange curves represent the best biophysical feature in each category, respectively: morphology (Diameter-RB in Table 1), optical phase (OPD-1 in Table 1), and optical loss (Absorption-2 in Table 1). The values represent average balanced accuracy among training datasets at the end of optimization. Clearly, the final achievable accuracy by multivariate classification is considerably higher than that of single features. (b) For each case, we show 5 ROC curves for different test datasets. The gray diagonal line shows results of random guess classification. Multivariate analysis based on TS-QPI images (blue curves) shows significant improvement in classification sensitivity and specificity. The fact that the classifiers remain almost unchanged during the five iterations of cross validation shows consistency and robustness of the classifiers. (c) To visualize the multivariate classification results, data points are depicted in the space of the first two PCA components.