by Chloé Benmussa ,Esther Sanfeliu,Anabel Martínez-Romero,Blanca González-Farré,Tomás Pascual,Joaquín Gavilá,Alona Levy-Jurgenson,Ariel Shamir,Fara Brasó-Maristany,Aleix Prat,Zohar Yakhini
Published: September 25, 2025
In this study, the authors show that latent feature representations derived from H&E-stained histopathological images preserve meaningful biological information across spatial scales — enabling robust tumor subtype classification and enhancing interpretability of deep learning models in computational pathology.


