Segmentation of magnetic resonance images using a combination of neural networks and active contour models

Abstract

Segmentation of medical images is very important for clinical research and diagnosis, leading to a requirement for robust automatic methods. This paper reports on the combined use of a neural network (a multilayer perceptron, MLP) and active contour model ('snake') to segment structures in magnetic resonance (MR) images. The perceptron is trained to produce a binary classification of each pixel as either a boundary or a non-boundary point. Subsequently, the resulting binary (edge-point) image forms the external energy function for a snake, used to link the candidate boundary points into a continuous, closed contour. We report here on the segmentation of the lungs from multiple MR slices of the torso; lung-specific constraints have been avoided to keep the technique as general as possible. In initial investigations, the inputs to the MLP were limited to normalised intensity values of the pixels from an (7 × 7) window scanned across the image. The use of spatial coordinates as additional inputs to the MLP is then shown to provide an improvement in segmentation performance as quantified using the effectiveness measure (a weighted product of precision and recall). Training sets were first developed using a lengthy iterative process. Thereafter, a novel cost function based on effectiveness is proposed for training that allows us to achieve dramatic improvements in segmentation performance, as well as faster, non-iterative selection of training examples. The classifications produced using this cost function were sufficiently good that the binary image produced by the MLP could be post-processed using an active contour model to provide an accurate segmentation of the lungs from the multiple slices in almost all cases, including unseen slices and subjects.

Multi-structure segmentation of multi-modal brain images using artificial neural networks

“…To overcome the limitations of the semi or fully automated segmentation method by using ANN, hybrid methods have also been proposed. Reddick et al suggested the combination of SOM and ANN for brain image segmentation [27], and Middleton et al proposed a different approach combining ANN and active contour model [21] . A hybrid method of wavelet with ANN has also been introduced [18].…”

Globally optimal co-segmentation of three-dimensional pulmonary 1H and hyperpolarized 3He MRI with spatial consistence prior

“…In another attempt, a modelmatching method was used to segment the lung parenchyma by registering individual images to a model created from manually segmented training dataset (Lelieveldt et al, 1999). Other approaches involved a combination of supervized neural network classifier and parametric active contour (Middleton and Damper, 2004). Another approach that incorporated merging of multiple parametric active contour within homogeneous regions (Ray et al, 2003) was also piloted to automatically segment the lung parenchyma.…”

“…For example, some of these approaches require substantial runtimes (Kirby et al, 2012a;Lelieveldt et al, 1999) while others were restricted to certain region of interest (Ray et al, 2003) or dependent on robust grey level thresholding (Sensakovic et al, 2006;Tustison et al, 2010;Woodhouse et al, 2005). Finally, other approaches required a substantial amount of user interaction (Kirby et al, 2012a), extensive training and frequent updating with careful expert manual segmentation (Middleton and Damper, 2004;Tustison et al, 2011).…”

“…Unfortunately however, because of intrinsic physiological and hardware variability (Robinson and Kreel, 1979), MR images cannot be processed based exclusively on signal intensities. The combined neural network and active contour method (Middleton and Damper, 2004) also required extensive training and updating with careful expert manual segmentation. The templatebased method (Tustison et al, 2011) was highly novel and used unbiased shape/intensity template construction, statistical lung shape descriptions and level set segmentation in 3D.…”

Multi-scale segmentation image analysis for the in-process monitoring of particle shape with batch crystallisers

“…Therefore here we make a brief overview of the methods available. Detailed reviews on image segmentation techniques can be found in the literature (Freixenet et al, 2002;Pal and Pal, 1993;Gonzalez and Woods, 1992;Deklerck et al, 1993;Freixenet, 2002;Russ, 2002), with the most popular methods being thresholding (Gommes et al, 2003;Otsu, 1979), split-and-merge (Manousakas et al, 1998;Deklerck et al, 1993), region growing (Lu et al, 2003;Hojjatoleslami and Kittler, 1998), clustering (Patino and Razaz, 2001), watershed segmentation (Razaz and Hayard, 2000), snakes (Middleton and Damper, 2004;Ngoi and Jia, 1999) and edge-detection-based methods (Rajab et al, 2004;Pavlidis and Liow, 1990). Some applications of these methods have been based on a colour segmentation scheme (Onyango and Marchant, 2003;Derganc et al, 2003), but nevertheless crystals generally show a translucent and colourless appearance and therefore the segmentation is better suited in grey scale.…”