NCSI: Semi-supervised Medical Image Synthesis with an Average Template

Status	Completed
Mentor Name	Xinlian Liu
Mentor's XSEDE Affiliation	Education Allocation, Campus Champion, Former Champion Fellow
Mentor Has Been in XSEDE Community	4-5 years
Project Title	Semi-supervised Medical Image Synthesis with an Average Template
Summary	Fake images generated by Deep Learning models such as GAN can be visually realistic trick people's eyes. However, they are not good enough to be used to train medical staff because 1) GAN synthesize images by approximating a target distribution. This leads to a pixelized picture with good texture information but inadequate shape information. 2) Often there are not enough training data set for a specific medical feature to start with. We propose to assist the model with a pre-generated standard image which is in turn produced by morphing/merging large amounts of inputs with lesser quality.
Job Description	1. Read journals and online literature to summarize recent development and identify potential useful tools and libraries. 2. Duplicate selected studies through paper reading. 3. Produce template images through merging training pictures. 4. Develop effective alignment algorithms. 5. Design and train neural nets to synthesize clinically realistic medical images. 6. Write reports and desiminate findings.
Computational Resources	PSC Bridges2-GPU. Our proposed work involves processing large amounts of imaging data and deep learning training as well as optimizations. These tasks require the GPU and a large memory partition.
Contribution to Community	This project will contribute to the understanding of a new approach of producing clinically-accurate medical iamges. It is also a training program to folster research skills for young researchers.
Position Type	Apprentice
Training Plan	1. Literature review; 2. Algorithms in Computer Vision; 3. Documentation and Code Maintainance; 4. Deep Learning Models implementation and training; 5. Scientific Writing
Student Prerequisites/Conditions/Qualifications
Duration	Semester
Start Date	01/15/2021
End Date	04/15/2021