Sites Inria

Version française


Sarah Alzieu - 7/06/2017

Digital Mammography Challenge : Therapixel wins the 1st round


With 5% of cancers still going undetected during the first X-ray reading, the "Digital Mammography Challenge" tackles a major public health issue. This year a start-up, Inria Therapixel, won the 1st round, thanks to a particularly effective algorithm. A great success for the company which turned its focus to artificial intelligence at the beginning of 2016.

The aim of the challenge, which was organised by around ten American organisations including IBM, Sage Bionetworks and Amazon, was twofold: to enable the improvement in the predictive precision of mammograms in the early detection of breast cancer and to reduce the number of unnecessary biopsies known as “false positives”.

Some 1,200 participants had a sizeable difficulty to overcome: the incredible quantity of available data (more than 640,000 images of anonymized, digital mammograms), paired with very limited calculation resources (barely two weeks in the Amazon cloud).

Finishing in first place at the end of four mega-competitive rounds, Therapixel’s algorithm, which was a compromise born of these constraints, raised hopes of a potentially significant improvement in various conditions.

Therapixel is a French start-up which was co-founded in 2013 by two researchers, Olivier Clatz and Pierre Fillard, and which is specialised in software development designed for medical imagery. As winner of the i-LAB 2013 and NETVA 2016 awards, it has been developing a new service to help interpret medical images since 2016. The process is based on artificial intelligence and aims to revolutionise diagnostic X-rays, in particular thanks to algorithms which use networks of convolutional neurons and which are now able to match the performance of the top experts in the field.