search

CBICA / CaPTk

Cancer Imaging Phenomics Toolkit (CaPTk) is a software platform to perform image analysis and predictive modeling tasks. Documentation: https://cbica.github.io/CaPTk
https://www.cbica.upenn.edu/captk
Other
182 stars 65 forks source link

Deep Learning based Tumor Segmentation does not work #497

Closed sarthakpati closed 5 years ago

sarthakpati commented 5 years ago

Describe the bug The brain tumor segmentation application using deep learning does not work.

To Reproduce Steps to reproduce the behavior:

  1. Load co-registered brain tumor subject (multi-modal with 4 modalities)
  2. Click on Applications > Brain Tumor Segmentation
  3. The python prompt starts up but exits with an error

Expected behavior The segmentation should work.

CaPTk Version 1.7.0.Beta

Desktop (please complete the following information):

sarthakpati commented 5 years ago

After conferring with Kostas (the algorithm developer behind DeepMedic), it was determined that the model getting shipped was an older version. This needed an update (will be present by default for the RC release). For users who have come across this thread, please use the file attached in this message and replace the contents in

brainTumorSegmentation.zip

MarkBergman-cbica commented 5 years ago

After conferring with Kostas (the algorithm developer behind DeepMedic), it was determined that the model getting shipped was an older version. This needed an update (will be present by default for the RC release). For users who have come across this thread, please use the file attached in this message and replace the contents in

Independent of DeepMedic, this is an example of the need for metadata (version number, date, etc) in every data file (model, CSV with pre-set features, etc) we distribute and a mechanism for each application to read and verify that meta-data (ie., model version 1.7 works with CaPTk 1.7--1.9). Bonus points for making the metadata human readable, from a literal sense (ie., a text-based comment, even in a binary file) and conceptual (ie., verbose metadata like "Model file foobar generated on 13 May 2019 has version 1.7a").

In the case of binary files, CaPTk should provide a "--about" flag (and GUI equivalent) to display metadata in models, binary masks, etc.