Niftynet meeting 4th december 2017

Attendance: Wenqi, Eli, Zach, Felix, Irme, Jieqing, Carole

Christmas release

• Model zoo ready but only 4 items
• Validation on the file ready
• Evaluation: almost all refactored from initial settings. Only output maps missing
• Need to prepare documentation out of the ReadMe:
  • Current situation: Lot of documentation but difficult to access
  • Solution Prepare a wiki/web based documentation page using the structure in which the README files are stored - Wenqi helped by Tom D

Model zoo extension

Need for a classification model. Example: Hemisphere location of tumour / Low grade vs High grade tumour. Zach to look into it using BRATS

Expected issue: Need to rethink the io for such application

Config file restructuration

In progress - Dzhoshkun + Carole working on it - Multiple files with details of interdependence

Keras

• Keras very simple to use for models.
• Need to include/allow for contributors to provide Keras-based models
• Need to make the best of available functionalities of Keras in order to strengthen NiftyNet
• Need to investigate feasibility of plugging Keras models in NiftyNet backend

Tensonflow dataset API queue will no longer be developed

Current problem with dataset: conversion of np array to tf tensor not allowing all functionalities

NiftyNet advertisement

Need to specify clearly the strengths/niche of NiftyNet when compared to Keras for instance

Value points of NiftyNet:

• IO Wrapper
• Medical Imaging focused Model zoo Need to work more on reproducibility -> advertised Need to advertise better after Christmas release