In an era of impressively high quality foundation models, AI teams need to work harder than ever to stand out. One of the best ways to differentiate is by building unique and proprietary datasets that can be used to update, fine-tune or evaluate foundation models for a targeted use case. 

Enabling AI teams to build these datasets - across data types, efficiently, and with unprecedented quality - is our top priority. That’s why today we’re announcing deeper support for DICOM and text labeling to streamline and accelerate labeling projects. 

‍

Introducing an updated DICOM labeling experience

Clinicians and drug developers often rely on medical images to inform critical diagnosis and treatment decisions, and to determine the efficacy of medicines. Enabling easier and more accurate interpretation of these information-rich images is an opportunity many AI teams are addressing - from Paige’s work with digital pathology, to Median Technology’s work with CT images. Today we’re excited to introduce updates to our end-to-end DICOM labeling experience that allows users to import DICOM files, create annotation tasks, view and label gold standards and enable mobile labeling contests right from within the Centaur Labs experience. 

‍

Integrating the OHIF viewer into our labeling workflow

We now offer a complete labeling solution for CT, MR and other complex modalities, by integrating the OHIF Viewer, a world class open source viewer developed by MGH, into the Centaur Labs platform. Users now have access to both the impressive capabilities of the OHIF Viewer, and to the privacy and security built into the Centaur Labs platform. Labelers and reviewers can open a DICOM, leverage the advanced visualization, measurement and segmentation capabilities available in the OHIF Viewer, apply a segmentation or review an existing one, and easily progress to the next image all from within the desktop experience. Project managers can upload studies directly to Centaur’s HIPAA and SOC2 Type II compliant platform, minimizing privacy and security risks. This more robust viewer enables SMEs to use familiar tooling to interpret medical images, so they can apply complex segmentations easily and accurately all while keeping their data private and secure, and their data labeling projects streamlined.

‍

‍

‍Windowing support on mobile 

We’re also enhancing our mobile DICOM viewer experience, making it easier to shift from manually labeling DICOMs on desktop, to scalably labeling with the global Centaur Labs network. Labelers can now - from their mobile device - move between all window levels by simply dragging their finger up, down or across the screen. This more dynamic viewer makes it easier to identify pathologies that are invisible in certain window levels,  so labelers can evaluate and classify or segment DICOMs accurately.

‍

‍

New text highlighting capabilities enable Named Entity Recognition and Linking 

Given the enormous corpus of biomedical and scientific documents - from unstructured patient notes and imaging reports, to chatbot conversations, to FDA filings and pharmaceutical press releases - many AI teams are building capabilities to both search these datasets more effectively,  generate domain-specific text by fine-tuning models with these datasets.  We’re excited to make our text highlighting capability generally available, to accelerate this exciting and important work.

‍

Highlight text seamlessly on mobile and desktop

Part of leveraging these datasets is highlighting and naming the entities important for your algorithm to identify. Perhaps you’re building a scientific literature search algorithm, and want to be able to find certain drug brands and drug targets. Those entities need to be highlighted in the training data you’re using to fine tune your search algorithm. 

With our new text highlighting capabilities, AI teams can set up highlighting tasks, label Gold Standards, and scale their text highlighting tasks to the Centaur Labs network. This new capability makes it possible to efficiently identify entities in text, making it easier for AI teams to build the text datasets needed to train and fine-tune their models. 

‍

‍

‍Learn more

• Connect with us to learn more, see a demo or discuss your use case. 
• These capabilities are generally available today to all customers globally.

‍