Unsupervised lesion detection on lung CTs with deep learning
Principal Investigator
Name
Andreas Baur
Degrees
Msc
Institution
University of Bern
Position Title
Student
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-459
Initial CDAS Request Approval
Dec 18, 2018
Title
Unsupervised lesion detection on lung CTs with deep learning
Summary
The aim of this project is to detect unhealthy lesions of any size at their locations. It shall help the physicians to recognize locations where tumour lesions could be.
The architecture of the project is an adversarial autoencoder which is trained on healthy subjects. The autoencoder will then reconstruct the input at the output. Healthy images are correct reconstructed, unhealthy images have differences at the locations of the lesions. To recognized a tumour lesion the output is subtracted from the input.
The architecture of the project is an adversarial autoencoder which is trained on healthy subjects. The autoencoder will then reconstruct the input at the output. Healthy images are correct reconstructed, unhealthy images have differences at the locations of the lesions. To recognized a tumour lesion the output is subtracted from the input.
Aims
- detect tumour lesions of any kind and size
- show the locations of the tumour lesions
- reach a sensitivity rate of 95%
Collaborators
this is a one person project