Detalle de: GeoTop: Advancing Image Classification with Geometric-Topological Analysis

We present GeoTop, a novel framework that synergises Topological Data Analysis and Lipschitz-Killing Curvatures to overcome limitations of single-modality feature extraction in biomedical multi-omics. By combining persistent homology’s capacity to identify robust topological signatures with Lipschitz-Killing Curvatures’ precision in quantifying morphological features, our approach achieves consistent accuracy improvements of 3-6% across diverse biological domains. The framework’s effectiveness is demonstrated through two key applications, i. e., skin lesion classification, where it reduces false positives/negatives by 15-18% compared to conventional methods, and plant signalling peptide identification, with F1-score: 0.98. GeoTop resolves the fundamental challenge of topological equivalence by incorporating geometric differentiators like perimeter and area profiles. Beyond its predictive performance, GeoTop offers three unique advantages; mathematical interpretability through persistence diagrams and curvature-based descriptors, computational efficiency for large datasets, and generalisability across biological scales - from macroscopic tumours to molecular structures. This work provides a principled approach for multi-omics integration, where complementary topological and geometric perspectives offer new insights into biological systems.