Interpretable Deep Learning of Endogenous Fluorophore Concentration and Entropy for Lung Tissue Classification

Combining endogenous fluorophore concentration and local histogram entropy improves normal versus cancerous lung tissue discrimination, reaching a mean ROC AUC of 98.4% ± 0.7% in five-fold cross-validation.

• Why it matters: Concentration and entropy capture different properties of fluorescence maps, but it is unclear whether using both yields complementary predictive information and which features drive decisions.
• What they did: Using chemometric fluorescence microscopy concentration maps of tryptophan, elastin, NADH, and FAD, the study compared concentration-only, entropy-only, and combined inputs with GAMI-Net using explicit main effects and pairwise interactions.
• The result: In the combined model, FAD entropy, NADH concentration, and NADH entropy were leading main effects, with learned log-odds showing U-shaped elastin-entropy and nonmonotonic tryptophan relationships.

AI summary of the abstract · may contain errors · not medical advice · read the paper.