Label-Free Classification of Bax/Bak Expressing vs. Double-Knockout Cells.

Abstract

We combine optical scatter imaging with principal component analysis (PCA) to classify apoptosis-competent Bax/Bak-expressing, and apoptosis resistant Bax/Bak-null immortalized baby mouse kidney cells. We apply PCA to 100 stacks each containing 236 dark-field cell images filtered with an optically implemented Gabor filter with period between 0.3 and 2.9μm. Each stack yields an "eigencell" image corresponding to the first principal component obtained at one of the 100 Gabor filter periods used. At each filter period, each cell image is multiplied by (projected onto) the eigencell image. A Feature Matrix consisting of 236×100 scalar values is thus constructed with significantly reduced dimension compared to the initial dataset. Utilizing this Feature Matrix, we implement a supervised linear discriminant analysis and classify successfully the Bax/Bak-expressing and Bax/Bak-null cells with 94.7% accuracy and an area under the curve (AUC) of 0.993. Applying a feature selection algorithm further reveals that the Gabor filter period ranges most significant for the classification correspond to both large (likely nuclear) features as well as small sized features (likely organelles present in the cytoplasm). Our results suggest that cells with a genetic defect in their apoptosis pathway can be differentiated from their normal counterparts by label-free multi-parametric optical scatter data.