DeepART: Medical Image collection, segmentation and anonymization for DEEP learning in Adaptive Radiation Therapy
The aim of the project DEEP learning in Adaptive Radiation Therapy (DeepART) is to take advantage of the impressive advances that the machine learning community has made recently, especially in deep learning approaches, to consider adaptive radiotherapy with innovative workflows to predict when replanning is relevant.
At present, most of the processing performed on the CT modality is carried out by ad-hoc methods. Recently, DNN, especially Convolutional Neural Networks (CNN), which are particularly adapted to images, are beginning to be used in this field. Nevertheless, the classification of 3D sequences, such as CT sequences, and the prediction of their short-term sequels with deep models remains to our knowledge an open scientific challenge, with important prospects in the medical field.
We propose to facilitate replanning by predicting the GTV evolution using CT images acquired during the treatment. To do so, we use Deep Neural Networks (DNN) adapted to sequence processing (able to deal with variable dimension signals) such as Long-Short Term Memory (LSTM), producing itself an image or a sequence of images in a generative mode, or combined with generative models such as Generative Adversarial Networks (GAN).
This project is partially funded by the MINMACS Région Normandie excellence label.
- LITIS (https://www.litislab.fr/)
- Centre François Baclesse (https://www.baclesse.fr/)
- Centre Henri Becquerel (https://www.becquerel.fr/)
- GREYC (https://www.greyc.fr/)