Classification des tumeurs intrahépatiques par méthode faiblement supervisée à partir de biopsies de routine



Main objectives

This project consists in developing a classification of primary liver cancers (CHC, iCCA, cCHC-CCA) using a weakly supervised method with the help of neural networks based on multi-scale data sources (3D radiology via CT-scan, 2D histology via histological biopsies).
The aim is to develop new artificial intelligence methods for better diagnosis of various primary liver cancers, and to make significant advances in patient care.


Primary liver cancers are the 4th leading cause of cancer deaths worldwide, with increasing incidence in most Western countries. They represent a heterogeneous group of tumors associated with distinct risk factors, clinical findings and imaging, histological and molecular features. Among these tumors, hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (iCCA) are by far the most common, representing the two extremes of the tumor spectrum. In between, combined hepato-cholangiocarcinoma (cCHC-CCA) is recognized as a rare tumor sharing features of both HCC and iCCA, whose prognosis remains unclear. While imaging and tumor biopsy separately achieve good results for the accurate diagnosis of HCC and iCCA, the diagnostic performance of cCHC-CCA based on a single modality is poor. Histological analysis represents the gold standard for the diagnosis of cHCC-CCA, but even from a histological point of view, this diagnosis can be very difficult, particularly in biopsy samples. Recently, we have shown that the combination of imaging and tumor biopsy can improve the diagnosis of cHCC-CCA, underlining the value of using multiscale imaging data.

Methodology/Technology used

The methodological approach is based on recent developments in artificial intelligence concerning weakly supervised learning from features extracted by pre-trained deep neural networks. In addition, one of the main novelties of the proposed study will be the development of a new approach to multi-scale data mining to take advantage of the two types of images available (CT-scan and histology).



Valérie Paradis () & Jean-Christophe Pesquet () & Aurélie Beaufrère () & Jules Gregory () & Nora Ouzir () & Kevin Mondet ()


AP-HP - Inria - CentraleSupelec

Useful links

Other projects



US Caractérisation et prédiction de la survenue de formes graves ou létales du COVID-19 à partir des données issues de l’EDS de l’AP-HP

Names of partners involved
AP-HP, Inria & Centrale Supélec

read more