Methods of Artificial Enlargement of the Training Set for Statistical Shape Models

Abstract

Due to the small size of training sets, statistical shape models often over-constrain the deformation in medical image segmentation. Hence, artificial enlargement of the training set has been proposed as a solution for the problem to increase the flexibility of the models. In this paper, different methods were evaluated to artificially enlarge a training set. Furthermore, the objectives were to study the effects of the size of the training set, to estimate the optimal number of deformation modes, to study the effects of different error sources, and to compare different deformation methods. The study was performed for a cardiac shape model consisting of ventricles, atria, and epicardium, and built from magnetic resonance (MR) volume images of 25 subjects. Both shape modeling and image segmentation accuracies were studied. The objectives were reached by utilizing different training sets and datasets, and two deformation methods. The evaluation proved that artificial enlargement of the training set improves both the modeling and segmentation accuracy. All but one enlargement techniques gave statistically significantly (p < 0.05) better segmentation results than the standard method without enlargement. The two best enlargement techniques were the nonrigid movement technique and the technique that combines principal component analysis (PCA) and finite element model (FEM). The optimal number of deformation modes was found to be near 100 modes in our application. The active shape model segmentation gave better segmentation accuracy than the one based on the simulated annealing optimization of the model weights.

Segmentation algorithms of subcortical brain structures on MRI for radiotherapy and radiosurgery: A survey

“…Statistical models (SM) have become widely used in the field of computer vision and medical image segmentation over the past decade [26,[45][46][47][48][49][50][73][74][75][76][77][78][79][80][81][82][83][84][85][86][87][88]. Basically, SMs use a priori shape information to learn the variation from a suitably annotated training set, and constrain the search space to only plausible instances defined by the trained model.…”

“…Other techniques focus on artificially enlarging the size of the training set. Koikkalainen et al [85] concluded that the two best enlargement techniques were the non-rigid movement technique and the technique that combines PCA and a finite element model.…”

Active shape model unleashed with multi-scale local appearance

“…This problem becomes more critical in medical image analysis due to inadequate training samples and varied pathologies. There have been strategies to relieve the over constraint, including loosening up the shape constraints [2,3], introducing additional flexibility [4], synthesising additional training samples [5], and more recently, modifying shape prior using manifold learning [6] or sparse composition [7]. In these methods there is no obvious a priori optimal parameters and it is time consuming to tune the parameters to specific applications.…”

A framework for the merging of pre-existing and correspondenceless 3D statistical shape models

“…To explicitly add shape variability to the models, a second popular approach consists of combining the statistical models with synthetic modes of variations (Lötjönen et al, 2005;Tölli et al, 2006;Koikkalainen et al, 2008). The types of deformation can be chosen to reflect expected variations, either through finite element analysis (Cootes et al, 1995;Wang and Staib, 2000) or heuristically (de Bruijne et al, 2003;Sclaroff and Isidoro, 2003).…”