Implementation of manual and smart semi-automatic brain lesion segmentation tools in SPINE

Context

Manual segmentations performed by experts, in medical imaging and in particular in brain imaging, are the gold standard for the evaluation of automatic segmentation methods. For the particular case of lesion segmentation, this manual task becomes time consuming, specially in cases where multiple lesions are to be detected and segmented. Semi-automatic methods help experts to label easy-to-segment cases and to approach challeging ones. In this project we would to implement the most common manual segmenation tools and more complex semi-automated methods. The tools will be implemente in the virtual laboratory SPINE. One additional feature of these approached, in particular the semi-automatic ones, is that the corrections made the experts can be used to improve them, eventually leading to the use of machine learning methods.

Basic segmentation tools

– Paint Brush

– Bezier curves (control points)

– Fill closed contour

Semi-automatic tools/methods

– Active contours 2D and 3D (snakes, level-sets)

Requirements

– Javascript

– Back-end: Node.js (hapi server), CouchDb

– Front-ent: React, Redux, and VTK.js

– Image processing

– Image processing (ITK c++ and/or js) is a plus

Contact:

Charles Guttmann, guttmann@bwh.harvard.edu
Alfredo Morales Pinzon, amoraleszpinon@bwh.harvard.edu
Andrzej Marciniak, amarciniak@bwh.harvard.edu

References:

– Level sets

– Bezier curves

– Tools:

– Jim’s tool

– Fastsurf

Opportunities