Project 4 - Image reconstruction using Deep Learning

# Project 4 - Image reconstruction using Deep Learning Lead: Imraj Singh Co-lead: Andrew Reader ### Description This group will look at deep learning techniques for PET reconstruction. In particular, we will look specifically at using a learned primal dual approach with simulated data. The simulated data will be of generated ellipses phantoms that will be forward projected and pre-corrected. This will constitute the training pairs (ground truth ellipse phantoms and associated pre-corrected sinograms). The learned primal dual approach will be simplified to reduce training costs, perhaps 3 primal and dual “nets”. ### Preparation #### Recommended videos: * [Basics of AI for PET Image Reconstruction](https://youtu.be/y-K-4Z_r3wg), Andrew Reader * [Pt 1: Brief Reconstruction Review and Machine Learning Approach, Linear Model and FBP](https://youtu.be/9HLDgGae40Y), Andrew Reader * [Pt 2: Deep Learning Convolution, Feature Hierarchy Abstraction, Convolutional Neural Networks](https://youtu.be/e8qout9FK18), Andrew Reader * [Pt 3: Direct deep learned image reconstruction (PET, MR, CT examples)](https://youtu.be/zA9yTqkt4nY), Andrew Reader * [Pt 4: Deep learned unrolled iterative image reconstruction](https://youtu.be/oQbMg35tAyY), Andrew Reader * [Simple PyTorch code to put deep learning into iterative image reconstruction (embeds a CNN in MLEM)](https://youtu.be/BXXLoVyAT0Q), Andrew Reader #### Recommended papers: * [Learned Primal-dual Reconstruction](https://arxiv.org/pdf/1707.06474.pdf), Jonas Adler and Ozan Oktem * [Learned Primal Dual Reconstruction for PET](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8707496/), Alessandro Guazzo and Massimiliano Colarieti-Tosti * ### Notebooks to be tackled during the conference:  ### Q&A # Back to main [Main page](https://hackmd.io/oNneaDUOQNK6XdDRmooITw?view#SIRF-training-school-PSMR-TBP-2022---Main)