DeformIt: Simulation of Ground Truth Data

Principal Investigator: Ghassan Hamarneh
Undergraduate Student: Preet Singh Jassi
MSc Student: Lisa Tang
website: DeformIt: Simulation of Ground Truth Data

The problem of scarcity of ground-truth, expert annotated medical image data is a serious one that impedes the training and validation of medical image analysis techniques. We develop algorithms for the automatic generation of large databases of annotated images from a single reference dataset and provide a web-based interface through which the users can upload a reference data set and download an arbitrary numbers of novel ground- truth data.


DeformIt allows the user to simulate novel data from a single reference dataset (an image, its segmentation, and its landmark points) capturing the exact anatomy on which the developed algorithms need to be trained or validated. We employ a physically- and statistically-based generative model to deform the single reference dataset.  Additionally, the reference dataset can be degraded by noise or non-uniformity intensities. 

See also related project VascuSynth on simulating tree-like structures. 


DeformIT now available at GITHUB https://github.com/erudianart/DeformIt/ 

Example results:

Reference and simulated images

Example 2D Simulation: (a) Reference image with different grid displacements (small arrows), (b) deformed checkerboard, (c) deformed images, (d) non-uniformity field, (e) ‘c+d’, (f) ‘e+noise’, (g) simulated segmentation (reference segmentation not shown

For more resutls see our paper below...



Ghassan Hamarneh, Preet Jassi, and Lisa Tang. Simulation of Ground-Truth Validation Data via Physically- and Statistically-based Warps. In Lecture Notes in Computer Science, Medical Image Computing and Computer-Assisted Intervention (MICCAI), pages 459-467, 2008.  


Site by Faculty of Applied Science web team, Simon Fraser University