GAS: A genetic atlas selection strategy in multi-atlas segmentation framework

Abstract

Multi-Atlas based Segmentation (MAS) algorithms have been successfully applied to many medical image segmentation tasks, but their success relies on a large number of atlases and good image registration performance. Choosing well-registered atlases for label fusion is vital for an accurate segmentation. This choice becomes even more crucial when the segmentation involves organs characterized by a high anatomical and pathological variability. In this paper, we propose a new genetic atlas selection strategy (GAS) that automatically chooses the best subset of atlases to be used for segmenting the target image, on the basis of both image similarity and segmentation overlap. More precisely, the key idea of GAS is that if two images are similar, the performances of an atlas for segmenting each image are similar. Since the ground truth of each atlas is known, GAS first selects a predefined number of similar images to the target, then, for each one of them, finds a near-optimal subset of atlases by means of a genetic algorithm. All these near-optimal subsets are then combined and used to segment the target image. GAS was tested on single-label and multi-label segmentation problems. In the first case, we considered the segmentation of both the whole prostate and of the left ventricle of the heart from magnetic

Validation of separate multi-atlases for auto segmentation of cardiac substructures in CT-scans acquired in deep inspiration breath hold and free breathing

“…Further improvement of atlas accuracy might be achieved by integration of an iterative atlas selection procedure or a genetic selection strategy to identify the best subset of atlases within the multi-atlas library [26,27]. Atlas generated contours can also be improved by correcting large errors [16,17].…”

An Evaluation of Atlas Selection Methods for Atlas-Based Automatic Segmentation in Radiotherapy Treatment Planning

“…Thus, early selection methods are inspired by intensity-based image similarity measures used in registration methods [52]. Proposed measures include mean absolute difference [33], sum-of-squared difference [30] (SSD), cross-correlation [35], [38], [50], [51] (CC) and normalized mutual information [23], [25], [27], [28], [31], [34], [37] (NMI). Following the same line of argument, it may be assumed that mapping a 'similar' atlas image to the patient image will require less deformation than mapping a less similar one.…”

“…The alternative approach which chooses the best combination of atlases is known in theory to perform better [59], but comes with an prohibitive computational cost for the size of the database considered in our study. A recent contribution suggested to reduce the complexity of the combinatorial selection by means of an optimization technique to select a 'near-optimal' subset [51]. Similarly, selection methods that learn surrogate measures of segmentation performance as in [42], [48] were also excluded because of the additional computational complexity associated with the need to train the system on an independent dataset.…”

A Survey on Nature-Inspired Medical Image Analysis: A Step Further in Biomedical Data Integration

“…Cagnoni et al [60] proposed an interactive contour-based delineation model for anatomical structures in tomographic medical images, where a GA drives the elastic-contour evolution. With reference to multi-atlas based segmentation, GAs were employed to select the near-optimal atlas sub-set combination to segment the current target image [61]. GAs are also effective for the selection of relevant and informative radiomics features [62], aiming at improving the discriminative power for subsequent classification tasks.…”