If you don’t find your question answered here, feel free to contact us at firstname.lastname@example.org.
ariadne was founded to catalyze manual image annotation, in particular neuron reconstruction efforts in the neuroscience community, by outsourcing image annotation tasks to full-time professionals. For most image annotation tasks in biology, 20–40 hours of training are necessary to become familiar with the task, software, and data. Large-scale image annotation pipelines that rely on students or volunteers usually suffer from a high turnover of workers, low individual throughput, and low or mediocre annotation quality, due to workers’ lack of long-term experience. Consequently, these pipelines involve a significant training and management overhead. ariadne minimizes overhead costs and maximizes annotation quality and throughput by maintaining a group of highly trained full-time annotators.
A team of expert project managers, all holding master’s degrees in life sciences, work together with experienced machine-learning engineers to solve your biomedical analysis problem. Our image annotators have many years of experience and are directly supervised by the project managers to ensure the highest quality possible.
The pricing of our high-throughput annotation service depends on the image quality, number of datasets and dataset size, and on the complexity and difficulty of the annotation task. We have many years’ experience in developing new image processing workflows that can be tailored to your analysis problems. The processing costs per Megavoxel typically drop significantly as the number of datasets and the dataset size increase.
Of course! Many of our workflows are easily adaptable to new segmentation targets and we love to tackle difficult biomedical image analysis problems. Please use our contact form to get in touch.
One of our services is end-to-end automated segmentation, including post-segmentation proofreading. First, we generate manual ground truth from scratch or use domain adaptation to create a powerful neural network model for segmentation. We then use an efficient iterative ground truth generation scheme to target the remaining low-quality regions. Finally, we polish the results with a manual proofreading step if necessary. Depending on the needs of our customers, we provide statistical analysis, quality scoring, 3D-reconstructions and videos of the data and deliver the results in a format that makes it easy for the client to pursue further research.
In addition to the segmentation of cells and their processes (e.g. neurites or filopodia), our team is experienced in analyzing various subcellular structures and organelles. Our capacities include workflows for both light microscopy and electron microscopy datasets.
We have extensive experience with a broad range of model organisms and tissues, ranging from cultured cells over muscle cells and stem-cell derived organoids, to the brain tissue of rodents and zebrafish.
Yes, we also deliver the manually annotated ground truth as well as manual annotation of the complete dataset, such as for neuron tracing projects.
In principle, the annotators can be trained to perform any kind of image annotation or labeling, and we have experience in annotating light microscopy data, serial block-face EM data, and ssTEM data. We are happy to adapt and develop new workflows for your annotation problems, should you have custom requirements. Please refer to our service section for a list of currently offered annotation tasks. If you have a different annotation task in mind, please do not hesitate to contact us. We continuously strive to extend our services and will be happy to set up a new workflow in collaboration with you.
Our current throughput is on the order of several thousands of person-hours per month. We can accommodate large projects by training additional annotators at any time.
Quality requirements usually vary with the annotation tasks and can be adjusted to customers’ individual requirements. Typical measures to ensure high accuracy involve independent, redundant annotation by different annotators; random control sampling by expert scientists; and the provision of regular, tailored feedback to the annotators.
Each annotator receives frequent feedback on his or her annotation quality and speed. Task-dependent benchmark annotation speeds are defined in close collaboration with the customer, in order to ensure both high quality and high efficiency.