Md

MD.ai develops software for medical imaging artificial intelligence and radiology workflows. Its Annotator product is a DICOM-native, browser-based platform for building and validating labeled medical imaging datasets, with features including an FDA 510(k)-cleared viewer, AI-assisted annotation, PHI detection and de-identification, developer APIs, and scaling for large projects. Annotations can be exported in formats such as JSON and DICOM SR, and users can search images by custom tags, mark up and comment on files, and use data directly in the application to train machine learning models.

A second product line focuses on clinical reporting, applying large language models to radiology reporting workflows. Described capabilities include automatic template selection, mapping of dictated key findings, impression generation, proofreading, automated billing code generation, patient-friendly audio messages, contextual AI chat, multilingual support and multi-device sync, with HL7/DICOM integration into EHR, HIS and RIS systems and both AI-driven and traditional reporting modes.

The company also publishes research and white papers with its team and collaborators, covering topics such as evaluation of GPT-4 for radiology report impression generation, evaluation of GPT-4V on detection of findings on chest radiographs, model fairness in image-based computer-aided diagnosis, and lessons from building multi-institution annotated datasets and hosting RSNA AI challenges.

Founding story

The company was started by physicians trained at Harvard, Duke and Columbia with the stated aim of accelerating the use of AI in medicine, particularly in medical imaging and associated software tools. George Shih, M.D., is identified as a co-founder.

Business model

MD.ai licenses its annotation platform and AI-assisted reporting software to healthcare and research organizations, with website engagement routed through demo requests and direct contact.

Traction

According to a Google Cloud case study, MD.ai's platform helped 1,300 teams process more than 30,000 images, and reduced project image management time from hours to seconds. The company states that thousands of doctors and collaborators have used the platform to create labeled datasets, deploy and validate models and build AI-driven clinical workflows, and that its software is used at top academic medical institutions and large pharmaceutical and healthcare companies. Stanford's AIMI Center provides the tool to its researchers.

Latest developments

Recent listed projects include the RSNA 2024 Lumbar Spine Degenerative Classification and RSNA 2023 Abdominal Trauma Detection competitions, and recent publications evaluate GPT-4 for radiology impression generation and GPT-4V for detecting findings on chest radiographs, the latter concluding the model is not ready for real-world diagnostic use in chest radiograph interpretation.