Encord DICOM Tool

Hello Product Hunt! We’re the makers of the Encord DICOM tool 👋 Here is a little background on why we built this: 🙋Problem The biggest bottleneck in healthcare & radiology AI right now is the difficulty involved in creating labeled datasets for AI models. Labeling medical imagery for training data requires thousands of hours of clinical time, which is a very expensive and time-consuming process. Existing DICOM annotation tools used in production are typically derived from open-source libraries such as ITK-Snap or 3D slicer. These are not purpose-built for AI data annotation, and they lack many features such as expert review workflows, automation features and model-assisted labeling, and label auditability. 💡 Solution That's why we spent a lot of time building out a new specialized DICOM annotation tool. We developed our tool in close collaboration with clinicians and healthcare data scientists to deliver expert functionality and an unparalleled user experience. We have tested our platform in use cases with Stanford Medicine, King’s College London, and the Memorial Sloan Kettering Cancer Center. The tool gives radiologists all the features they need and expect to make efficient and precise annotations on medical data: -Support for CT, X-ray, MRI, ultrasound, mammograms and more. -Coronal, axial and sagittal 3D views - all in a single viewer. -Hanging protocols - measuring tools, window widths & levels defined by Hounsfield units. -Fully auditable labels for FDA approval - keeping an audit trail of every single label. -Rendering of 20,000+ pixel intensities vs. 256 in a regular browser canvas. -Powerful automation features for model-produced annotations & interpolation. -Configurable expert review workflows for QA & clinician review. -Dashboards of annotator throughput and quality. With this tool, we both accelerate the workflow for people building out medically-focused AI training datasets and increase the quality and auditability of these datasets. Combined with our automated micro-model approach, we hope to remove the training data bottleneck hindering medical AI. Our platform has helped build datasets for precancerous polyps, detecting tumors in brain scans, classifying ultrasounds, and more. We are proud of our new tool and hope that people in the PH community will find it useful in building out the next generation of medical AI models! To get started, come talk to us! We look forward to hearing your feedback 🚀

Here's a great introduction to how this tool works to efficiently label imagery, using Batman as the test case, by @eric_landau !

Great work by @rad_ploshtakov1, @eric_landau, @ulsha and the Encord team!