CMIC Seminar: Eli Gibson and Aurobrata Ghosh

Speaker: Eli Gibson and Aurobrata Ghosh
UCL Contact: Dominique Drai (Visitors from outside UCL please email in advance).
Date/Time: 08 Jul 15, 13:00 - 14:00
Venue: Roberts 421

Abstract

Eli Gibson

Title: Statistical power in image segmentation: relating sample size to reference standard quality

Abstract:

Ideal reference standards for comparing segmentation algorithms balance trade-offs between the data set size, the costs of reference standard creation and the resulting accuracy. As reference standard quality impacts the likelihood of detecting significant improvements (i.e. the statistical power), we derived a sample size formula for segmentation accuracy comparison using an imperfect reference standard. We expressed this formula as a function of algorithm performance and reference standard quality (e.g. measured with a high quality reference standard on pilot data) to reveal the relationship between reference standard quality and statistical power, addressing key study design questions: (1) How many validation images are needed to compare segmentation algorithms? (2) How accurate should the reference standard be? This talk will offer some key insights into the derivation of the formula, and briefly explore a case study showing the practical use of the formula, using the PROMISE12 prostate segmentation data set.

Aurobrata Ghosh

Title: Invariants and Scalar Indices of a 4th order tensor: Building blocks for new "biomarkers" from HARDI

Abstract:

Model free approaches for HARDI are hugely popular in dMRI and bases such as Spherical Harmonics (SH) and Cartesian tensors have been widely used. In this talk, I will present some of my previous work done at Inria, where I will present methods to systematically compute all possible invariants to rotation of a 4th order tensor (equivalently SHs). Since these bases are used to represent numerous HARDI descriptors (ADC, ODF, FOD, etc.), the proposed methods can be used to capture the pure shape characteristics from a wide range of local diffusion (spherical) functions. First, we establish the total count of possible invariants,then attempt to recover the functionally smallest and complete set of invariants. Similar to DTI "biomarker" can these invariants be used to design HARDI scalar indices in the future?