Limits...
Automated detection and segmentation of synaptic contacts in nearly isotropic serial electron microscopy images.

Kreshuk A, Straehle CN, Sommer C, Koethe U, Cantoni M, Knott G, Hamprecht FA - PLoS ONE (2011)

Bottom Line: The statistical learning is performed on geometrical features of 3D neighborhoods of each voxel and can fully exploit the high z-resolution of the data.On a quantitative validation dataset of 111 synapses in 409 images of 1948×1342 pixels with manual annotations by three independent experts the error rate of the algorithm was found to be comparable to that of the experts (0.92 recall at 0.89 precision).Our software offers a convenient interface for labeling the training data and the possibility to visualize and proofread the results in 3D.

View Article: PubMed Central - PubMed

Affiliation: Interdisciplinary Center for Scientific Computing, University of Heidelberg, Heidelberg, Germany.

ABSTRACT
We describe a protocol for fully automated detection and segmentation of asymmetric, presumed excitatory, synapses in serial electron microscopy images of the adult mammalian cerebral cortex, taken with the focused ion beam, scanning electron microscope (FIB/SEM). The procedure is based on interactive machine learning and only requires a few labeled synapses for training. The statistical learning is performed on geometrical features of 3D neighborhoods of each voxel and can fully exploit the high z-resolution of the data. On a quantitative validation dataset of 111 synapses in 409 images of 1948×1342 pixels with manual annotations by three independent experts the error rate of the algorithm was found to be comparable to that of the experts (0.92 recall at 0.89 precision). Our software offers a convenient interface for labeling the training data and the possibility to visualize and proofread the results in 3D. The source code, the test dataset and the ground truth annotation are freely available on the website http://www.ilastik.org/synapse-detection.

Show MeSH

Related in: MedlinePlus

User labels and algorithm predictions.Top row: the complete set of user annotations for the first training set (20 brush strokes in total), with yellow labels for synapses, red for membranes, green for the rest. Bottom row: raw data and algorithm predictions on two other slices in the first training set. In black circles: some unlabeled synapses and their probability maps. The color intensity corresponds to the certainty in the prediction, predictions for green class are omitted for clarity.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3198725&req=5

pone-0024899-g001: User labels and algorithm predictions.Top row: the complete set of user annotations for the first training set (20 brush strokes in total), with yellow labels for synapses, red for membranes, green for the rest. Bottom row: raw data and algorithm predictions on two other slices in the first training set. In black circles: some unlabeled synapses and their probability maps. The color intensity corresponds to the certainty in the prediction, predictions for green class are omitted for clarity.

Mentions: To quantitatively assess the algorithm performance and its stability with regard to the training data, four training sets were created from images acquired in the same experiment, but not overlapping with the test set. The four training sets were located in different parts of the image stack and contained approximately the same number of voxel labels. For each training set, 2–3 synapses were labeled, and for each of those synapses it was sufficient to only label it in one of the slices. Adding more labels did not improve the classification performance, as long as the already labeled set represented the data well, which can be judged, for example, by looking at the current algorithm predictions for some non-labeled synapses (Fig. 1, bottom row). Although the software can discriminate an arbitrary number of categories, we found three-class labeling of synapses vs. membranes vs. the rest of the tissue to produce the best results. One can also use a binary setup with synapses vs. the rest, but then the labeler has to take extra care to annotate enough membrane voxels to obtain a representative sample of the background. Adding more classes, for example, for the mitochondria, did not help the classification. Our first training set is illustrated in Fig. 1 and a performance comparison for the different training sets is shown in Fig. 2A.


Automated detection and segmentation of synaptic contacts in nearly isotropic serial electron microscopy images.

Kreshuk A, Straehle CN, Sommer C, Koethe U, Cantoni M, Knott G, Hamprecht FA - PLoS ONE (2011)

User labels and algorithm predictions.Top row: the complete set of user annotations for the first training set (20 brush strokes in total), with yellow labels for synapses, red for membranes, green for the rest. Bottom row: raw data and algorithm predictions on two other slices in the first training set. In black circles: some unlabeled synapses and their probability maps. The color intensity corresponds to the certainty in the prediction, predictions for green class are omitted for clarity.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3198725&req=5

pone-0024899-g001: User labels and algorithm predictions.Top row: the complete set of user annotations for the first training set (20 brush strokes in total), with yellow labels for synapses, red for membranes, green for the rest. Bottom row: raw data and algorithm predictions on two other slices in the first training set. In black circles: some unlabeled synapses and their probability maps. The color intensity corresponds to the certainty in the prediction, predictions for green class are omitted for clarity.
Mentions: To quantitatively assess the algorithm performance and its stability with regard to the training data, four training sets were created from images acquired in the same experiment, but not overlapping with the test set. The four training sets were located in different parts of the image stack and contained approximately the same number of voxel labels. For each training set, 2–3 synapses were labeled, and for each of those synapses it was sufficient to only label it in one of the slices. Adding more labels did not improve the classification performance, as long as the already labeled set represented the data well, which can be judged, for example, by looking at the current algorithm predictions for some non-labeled synapses (Fig. 1, bottom row). Although the software can discriminate an arbitrary number of categories, we found three-class labeling of synapses vs. membranes vs. the rest of the tissue to produce the best results. One can also use a binary setup with synapses vs. the rest, but then the labeler has to take extra care to annotate enough membrane voxels to obtain a representative sample of the background. Adding more classes, for example, for the mitochondria, did not help the classification. Our first training set is illustrated in Fig. 1 and a performance comparison for the different training sets is shown in Fig. 2A.

Bottom Line: The statistical learning is performed on geometrical features of 3D neighborhoods of each voxel and can fully exploit the high z-resolution of the data.On a quantitative validation dataset of 111 synapses in 409 images of 1948×1342 pixels with manual annotations by three independent experts the error rate of the algorithm was found to be comparable to that of the experts (0.92 recall at 0.89 precision).Our software offers a convenient interface for labeling the training data and the possibility to visualize and proofread the results in 3D.

View Article: PubMed Central - PubMed

Affiliation: Interdisciplinary Center for Scientific Computing, University of Heidelberg, Heidelberg, Germany.

ABSTRACT
We describe a protocol for fully automated detection and segmentation of asymmetric, presumed excitatory, synapses in serial electron microscopy images of the adult mammalian cerebral cortex, taken with the focused ion beam, scanning electron microscope (FIB/SEM). The procedure is based on interactive machine learning and only requires a few labeled synapses for training. The statistical learning is performed on geometrical features of 3D neighborhoods of each voxel and can fully exploit the high z-resolution of the data. On a quantitative validation dataset of 111 synapses in 409 images of 1948×1342 pixels with manual annotations by three independent experts the error rate of the algorithm was found to be comparable to that of the experts (0.92 recall at 0.89 precision). Our software offers a convenient interface for labeling the training data and the possibility to visualize and proofread the results in 3D. The source code, the test dataset and the ground truth annotation are freely available on the website http://www.ilastik.org/synapse-detection.

Show MeSH
Related in: MedlinePlus