New in-situ and in-vivo diagnostic technologies that can help clinicians in the detection and discrimination of hyperplastic and neoplastic polyps and assist them in the polyp resection process are demanded. To this end, in the PICCOLO project we have developed a deep learning model that is capable of recognizing images of malignant neoplastic lesions (adenocarcinomas) and distinguish them from images of healthy, hyperplastic or benign neoplastic (adenomas) tissues, by autonomously extracting imaging biomarkers present in human colon tissue images acquired with a multiphoton microscope.

The baseline algorithm, which has been trained, validated and tested against a preliminary version of the project human model ex-vivo MPM database, gets 80.11±12.52% sensitivity and 85.76±9.54% specificity when classifying images that correspond to tissue tiles of 511x511 µm². The results improve with the proposed spatially coherent predictions (SCP) scheme, that considers several adjacent tiles for diagnosis prediction, achieving 82.28±15.75% sensitivity and 91.14±8.14% specificity.

These results confirm that the combination of this novel imaging technology together with deep learning algorithms leads the way to perform real-time optical biopsies for in-vivo diagnosis. Besides, it can be combined with other complementary novel imaging approaches as the OCT Classifier for improved diagnosis.

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

CONTACT

Elena Terradillos

Tecnalia Research & Innovation (TECNALIA)

www.tecnalia.com

elena.terradillos@tecnalia.com