With the recent advent of digital image acquisition technologies, scientists routinely generate larger and larger amounts of imaging data related to biological processes or diseases e.g. in neuroscience, anatomical pathology, drug discovery, toxicology, and disease studies.
Projects leading to terabytes of imaging data are becoming usual, e.g. when experimental studies rely on whole-slide virtual microscopy, high-content screening, automated volume electron microscopy, ...
Science being collaborative, needs for remote visualization and collaborative annotation are there, but common tools are hardly able to support such datasets in a distributed and collaborative way. Moreover, human interpretration of such datasets is time-consuming and somehow subjective, so this stresses the need for replicable and generic computational methods to automate the extraction of useful, quantitative, information from these images.
Towards that goal, there is a strong need for intuitive and efficient tools to foster collaboration between life science researchers and computer scientists who need large and realistic training datasets to develop effective machine learning and computer vision methods.
CYTOMINE is a rich internet application for remote visualization, collaborative & semantic annotation, and (semi-)automated analysis of high-resolution (multi-gigapixel) images using recent web development, and machine learning techniques.
The platform has been designed to faciliate accessibility, curation, and dissemination of imaging data, and to be widely applicable and extensible. Although it has been motivated by biomedical applications, it can be used in other application domains where there are very large images (small ones also). Also, although it has been designed for remote collaboration, it can be installed and used on a local computer for small-scale studies. For more details, you can read our paper (Marée et al., Bioinformatics 2016) DOI: 10.1093/bioinformatics/btw013.
It relies on different technologies (but can be extended to use other technologies):
- Whole-slide scanners (or other imaging equipments) to convert glass slides (or other physical samples) into high-resolution (gigapixel) images
- Modern web development tools & libraries so that the main Cytomine functionalities are available through a regular web browser
- Image analysis and generic machine learning algorithms
- High-performance computing and mass storage equipments (although a Cytomine server can also be run on a laptop/desktop)
The overall architecture of the platform is depicted below, where end-users (e.g. biologists) can use a regular web browser to access from anywhere in the world all their images and data (annotations, results) stored in centralized server databases.
Similarly, computer/data scientists can access (import & export) data using third-party softwares (written e.g., but not limited to, in Python or Java) through a RESTful API:
End-users (e.g. biologists, biomedical researchers) and computer scientists (e.g. image analysis developers) can use this software together to develop novel quantification workflows, e.g. we used it for tumoral tissue quantification in whole H&E slides (see our (Marée et al., Bioinformatics 2016) DOI: 10.1093/bioinformatics/btw013 paper and other papers citing Cytomine to learn more about other applications)