Zero-Shot Learning to Simultaneously Recognize and Localize Cancer in PET images
“Zero-Shot Learning” aims at simultaneously recognizing and locating object instances belonging to novel categories without any training examples . This method is an end-to-end deep network that jointly models the interplay between visual and semantic domain information. To overcome the noise in the automatically derived semantic descriptions, the concept of meta-classes is used to design an original loss function that achieves synergy between max-margin class separation and semantic space clustering.
The objective of this master is to detect and localize cancers from PET (positron emission tomography) images using this new technique “Zero-Shot Learning”. The health patients without cancer are used for learning. New cancers are detected as new class. This method can overcome the challenge of small size of data in medicine field.
The objective of this master's internship is to detect and localize cancers from PET (positron emission tomography) images using this new “Zero-Shot Learning” technique. Healthy patients without cancer are used for learning. For cancer patients, cancers that are visible on PET images can be detected as a new class and be located on the images. This method can overcome the challenge of the small data size in medicine and detect the onset of cancer early.
1.The intern should have strong image processing and Python programming skills.
2.This master will be supervised by Su Ruan ( LITIS Rouen, <email@example.com>) and Abderrahim El Moataz Billah (GREYC Caen, <firstname.lastname@example.org>) in collaboration with the Centre de lutte contre le cancer Henri Becquere in Rouen.
3.The internship will take place in the Quantif team of the LITIS laboratory, Faculty of Medicine, Rouen
(c) GdR 720 ISIS - CNRS - 2011-2020.