as shuffle radius is increased, errors decrease in both cases
We can expect shuffle distance of size 7x7 to be the most sensitive
Generalisation and Error Bars
Generalisation ability plot with error bars re-calculated in accordance with last meeting
Specificity and Error Bars
Specificity plot with error bars re-calculated in accordance with last meeting
Sensitivity Plots - Model-Based
Sensitivity is shown for a variety of model-based evaluations
Must take a second look at the errors being computed (take with a grain of salt)
It is suggested that the errors should be ignored for the time being
Can apply the code to both overlap- and model-based measures, plotting them in tandem when/if necessary
Sensitivity of Generalisation with Error Bars
Sensitivity of Generalisation ability
Sensitivity of Specificity with Error Bars
Sensitivity of Specificity ability
Sensitivity Plots - Overlap-Based
Sensitivity for overlap-based evaluations
No errors are shown yet
Measures were more sensitive not quite where we had expected
Overlap is formulated to have opposite trends to that of shuffle distance (i.e. error, discrepancy)
Hence, ought to be looked at 'in reverse'
Sensitivity of Overlap without Error Bars
Sensitivity of overlap-based evaluation
Papers of Interest/Relevance
Most relevant ones appear in a recent issue of TMI:
Zagrodsky, V. Walimbe, V. Castro-Pareja, C.R. Qin, J.X. Song, J.-M. Shekhar, R. - Registration-Assisted Segmentation of Real-Time 3-D Echocardiographic Data Using Deformable Models
Beichel, R. Bischof, H. Leberl, F. Sonka, M. - Robust Active Appearance Models and Their Application to Medical Image
Analysis
VandeKraats, E.B. Penney, G.P. Tomazevic, D. VanWalsum, T. Niessen, W.J. - Standardized Evaluation Methodology for 2-D/3-D Registration