|
|
Planning with slides and videosProgramMonday April 1508h30 - 10h00 Welcome coffee and registration 10h00 - 10h30 Introduction: Program of the school 10h30 - 12h30 Basics of machine learning - Part 1, Part 2, Part 3 and Part 4 Video P1 Basic Learning 1 Carole Lartizien and Rémi Emonet Video P2 Basic Learning 2 - Nicolas Duchateau 16h30 - 17h30 Learning with less labels in medical image analysis (canceled) 17h30 - 18h30 Round table: Needs of clinicians, researchers and industrials Moderator: Paul De Brem Video P3 Table Ronde: Needs of clinicians, researchers and industrials 18h30 - 19h30 Conference - Demystification of AI-driven medical image interpretation: past, present and future Video P4 Demystification of IA - Benoit Gallix 20h00 - 22h00 Convivial evening on the campus at the Supr'M Tuesday April 169h00 - 10h30 Basics in deep learning - Part 1 and Part-2: Perceptron, Stochastic Gradient Descent, Learning rate, Logistic regression, Cross-entropy, Multi-layer Perceptron, Activation functions, Multi-Class Perceptron Video P5 Basic Deep Learning Pierre Marc Jodoin Part I 10h30 - 11h00 Coffee break 11h00 - 12h30 Basics in deep learning - Part 2: Forward pass and Back propagation, Batch size, Feature maps (architecture selection), Convolution neural nets (CNN) - Properties, advantages, applications, Pooling functions Video P6 Basic Deep Learning Pierre Marc Jodoin Part II 12h30 - 14h00 Lunch break - Sogeres restaurant 14h00 - 18h00 Hands-on session: classification from cardiac shape data Format: python/ jupyter notebook Data: set of 2D myocardial shapes from 101 subjects at end-diastole and end-systole, partially pre-processed to manually extract features of interest. Labels are either healthy or diseased, and are unknown for 30 subjects. Content: test classifiers from scikit-learn to label the unknown cases. We will carefully evaluate the learning performance keeping an eye on the initial data and the extracted features. 20h00 - 22h00 Poster session open to everyone (participants and speakers) - buffet Wednesday April 179h00 - 10h30 Advanced concepts in deep learning Video P7 Advanced Deep Learning Christian Desrosiers 10h30 - 11h00 Coffee break 11h00 - 12h30 Deep learning in cardiovascular applications - second part Video P8 Deep Learning Cardiovascular Ozan Oktay 12h30 - 14h00 Lunch break - Sogeres restaurant 14h00 - 18h00 Hands-on session: Automatic segmentation of 2D echocardiographic images by deep learning Format: python/ Jupyter Notebook Data: database from 500 patients of 2D sequences (from two different orientations) named CAMUS database. This database is public and we will be able to use it (and promote it) during the school. Content: implement step by step a convolutional neural network (U-net) to automatically segment 2D echocardiographic sequences (both 2 chamber and 4 chamber view orientations). 20h00 - 22h00 Evening in Lyon Thursday April 189h00 - 10h30 Weakly supervised and disentangling Video P9 Weakly Supervised Nicolas Thome 10h30 - 11h00 Coffee break 11h00 - 12h30 Generative adversarial networks for medical imaging Video P10 Generative Networks Medical Anirban Mukhopadhyay 12h30 - 14h00 Lunch break - Sogeres restaurant 14h00 - 18h00 Hands-on session: Medical image reconstruction by deep learning Format: python/ Jupyter Notebook Data: 1) training and validation sets generated using a physical ultrasound simulator. 2) test set, from the PICMUS challenge, composed of in-vitro and in-vivo data acquired using the Verasonics® Vantage™ Research System. Content: Adapt the popular U-Net architecture for the purpose of image reconstruction. Learn to reconstruct high-quality ultrasound images from data acquired at high frame rate with poor image quality 20h00 - 22h00 Gala dinner on a boat Friday April 1909h30 - 11h00 Deep learning in neuroimaging 11h00 - 11h30 Coffee break 11h30 - 12h20 Conclusion and discussion with particpants 12h20 - 14h00 Lunch break -Sogeres restaurant 14h00 - 18h00 Hands-on session: which programming for deep learning? Format: python/ Jupyter Notebook Data: 2D images from MRIs and CT Content: compare same neural network implemented on different frameworks (pyTorch/TensorFlow) on basic classification task. Back home with good practices and compare results! |
| Personnes connectées : 1 |
|
