MONAI Label for AI-Assisted Medical Image Annotation

Medical imaging is a critical frontier in healthcare, enabling deeper understanding of anatomical structures and pathological conditions. It’s crucial for innovations in disease diagnosis, treatment planning, and the development of medical interventions. Traditionally, the complexity of medical images and the time-consuming process of manual analysis posed challenges to medical imaging analysis. Now, these are being addressed by integrating AI and accelerated computing for medical imaging analysis, such as the NVIDIA NIM™ VISTA-3D.

AI-assisted annotation in medical imaging has revolutionized our approach to medical imaging by enabling rapid and accurate interpretation of complex scans, often with sub-millimeter precision. This not only saves time and resources, but also enhances understanding of disease progression and treatment planning.

Project MONAI is an advanced AI platform for medical imaging that includes MONAI Label for AI-assisted annotation. MONAI utilizes AI methods to:

• Select the most informative data to be annotated
• Provide initial annotation estimates that can be quickly edited by experts to create high-quality truths
• Support interactive editing using a variety of simple user interactions such as squiggles and seeds
• Use online, active learning methods for continuously improving the primary AI model as new cases are annotated

MONAI Label has been integrated with multiple interactive annotation tools such as 3D Slicer, OHIF, and CVAT for 2D and 3D image and video annotation. Utilizing a client-server architecture that can be easily adapted to edge and cloud applications, MONAI Label can be accelerated using NVIDIA hardware to:

• Accurately analyze and annotate medical images, from routine diagnostic scans to complex research datasets, improving efficiency and precision in image interpretation
• Access state-of-the-art AI models including the VISTA-3D NIM, a domain-specialized interactive foundation model, to provide initial labels, requiring minimal editing to achieve highly accurate models for focal or whole-body segmentations, disease identification, or surgical event recognition
• Develop and refine AI-driven medical imaging workflows tailored for a wide range of clinical applications, using flexible deployment options across cloud or on-prem to meet diverse research and healthcare needs