Haematoxylin-Eosin (H&E) image classifier allows malignancy detection over histopathology images. Our results show a balanced accuracy of 0.87 over single tissue tiles. These results were increased up to 0.95 when addressing representative tiles from the tissue slide. These results confirm that the application of deep learning algorithms lay the foundations for the development of real-time optical biopsy applications.

Traditional deep learning methods are not tractable for how they ended up to that decision.  We introduce the concept of uncertainty range where the network is not confident on its own prediction. Increasing this uncertainty range deals to higher performance on the network output which allows also to determine if additional confirmation analysis is needed.

The details of this output will be openly accessible through publications in scientific journals at the scientific community´s disposal.

CONTACT

Artzai Picón

Tecnalia Research & Innovation (TECNALIA)

www.tecnalia.com

artzai.picon@tecnalia.com