Edit this page

NA-MIC Project Weeks

Back to Projects List

Decision Support for End-Stage Liver Disease Transplants

Key Investigators

Project Description

The waiting list for liver transplants is ordered according to a score based on multiple criteria. However, qualitative parts describing the quality of life (encephalopathy, coma, ascities, GI or variceal bleeds) have been excluded from the score because they were subjective too easy to “game” to move closer to transplant. These scores are important, though, so we would like to introduce a quantitative, image-based scoring in order to

We are using Partner’s image database for a corpus of imaging data (liver disease patients and controls with the same scan protocol but no liver disease).


  1. Understand standard/established clinical scores and effects/representation in medical imaging.
  2. Familiarize with the data from this study.
  3. Review and discuss current literature/ feature extraction approaches.
  4. This is kind of a Project Kick-Off: Create a work plan how to approach this problem also beyond the scope of the project week.

Approach and Plan

  1. Discuss features and feature extraction.
  2. Radiomics is mostly done on CT, not so much on MR. Applicability of pyradiomics features?


  1. We had a first team meeting to bring together computer scientists and clinicians.
  2. Dr. Wall reviewed her progress in selecting a small set of optimal diseased and control patients. This process has been challenging because many people with liver disease have had surgery or tumor ablation that changes the liver morphology. It is also not possible to select only patients on 3T scanner before BWH began using EPIC (2015).
  3. Alireza Ziaei, Raul San Jose, and Randy Gollub are assisting with RPDR querying and image retrieval.
  4. Jennifer worked on CITI training for IRB clearance to access the data. And talked with experts using PyRadiomics on MRI Data and their approaches on evaluating features (Michael Schwier and Joost van Griethuysen).

Next Steps

  1. Lock down the image queying and retrieval pipeline.
  2. Get deidentified data to University of Bremen team.
  3. Think hard about segmentation, machine learning, and analysis techniques for the data.

Background and References