FIGURE 5.

Metabolic activities of breast cancer cells are predicted from autofluorescence lifetime images. CNN models trained on autofluorescence lifetime images achieved approximately 95% accuracy in classifying glycolytic and oxidative cancer cells. (A) Predicting MCF7 metabolism as glycolysis inhibition or OXPHOS inhibition from autofluorescence lifetime images with CNN. The LeNet CNN models were trained with different fluorescence lifetime component images (NADH τ ₁ , NADH τ ₂ , NADH α ₁ , NADH τ _m , NADH intensity, FAD intensity), scale bar = 7 μm. (B) AUC ROC curves and accuracy for the test dataset (n = 1,520) of each classifier built to predict metabolism as glycolysis inhibition or OXPHOS inhibition. RFT: random forest tree; SVM: support vector machine; QDA: quadratic discriminant analysis.