A Comparative Study of Feature Descriptors for Mitochondria and Synapse Segmentation

Abstract

Abstract-Full understanding of the architecture of the brain is a long term goal of neuroscience. To achieve it, advanced image processing tools are required, that automate the the analysis and reconstruction of brain structures. Synapses and mitochondria are two prominent structures with neurological interest for which various automated image segmentation approaches have been recently proposed. In this work we present a comparative study of several image feature descriptors used for the segmentation of synapses and mitochondria in stacks of electron microscopy images.

Multi-class Boosting for Imbalanced Data

“…In our experiment we used an image stack obtained from the somatosensory cortex of a rat, with a resolution of 3.686 µ,m per pixel. The thickness of each layer is 20 µ,m [12]. From this data set we collected a training set composed of 10,000 background, 4000 mitochondria and 1000 synapse data and a testing set with 20,000 data per class.…”

“…In this section we use the BAdaCost algorithm to label pixels in these images as mitochondria, synapse and background, and compare the results with those achieved by the AdaC2.Ml algorithm. Following [12], we apply to each image in the stack a set of linear Gaussian filters at different scales to compute zero, first and second order derivatives. For each pixel we get a vector of responses In this experiment we compare BAdaCost (with Oil costs), AdaC2.Ml and BAdaCost (with the imbalanced cost matrix described in Sect.…”

Image segmentation with boosted multi-class classifiers

“…We experimentally show, that by selecting the adequate set of features, it is possible to achieve better performance than the state-of-the-art approaches. Our comparative study for feature selection for brain structures segmentation was presented in [16]. The scale selection with PIBoost algorithm and its experiments was presented in [15].…”

“…The elements of the selected channels can be seen in figure4.2. We choose GRIMS because they are an excellent descriptor for segmenting mitochondria and synapses [16,74] as seen in Chapter 3. Since vesicles are a good indicator of the existence of synapses in the vicinity (see the raw image in Figure 4.2), we also include an elliptical descriptor that provides contextual information related to the existence of vesicles.…”

“…Classification with high disproportion of data instances between classes is very common in machine learning since many real-world problems have a level of inherent asymmetry with different class priors. This asymmetry can be easily seen in detection [30] and segmentation [16,7] problems where the background data tends to be more numerous than the foreground classes. Asymmetric problems are different than imbalanced problems, as the later classification problems happen when, in the training set there is much less data in some classes than others in spite of all the classes having equal prior probabilities.…”