MICCAI 2021 Daily – Thursday

19 DAILY MICCAI Thursday For improved performance for a classical medical image analysis problem, MONAI v0.7 running on a V100 GPU can achieve training convergence to a validation mean dice of 0.95 within 1 minute on the spleen segmentation task of MSD challenge. Compared to MONAI v0.6, the convergence time is 20x faster! Learn More MONAI 0.7 also includes many new network models, such as Transchex which consists of vision, language, and mixed-modality transformer layers for processing chest X-ray and their corresponding radiological reports within a unified framework. MONAI’s research contributions also include a state-of-the- art implementation of UNETR, a transformer-based model for volumetric (3D) multi-organ segmentation tasks using the BTCV challenge dataset. Tutorial MONAI core’s goal is to power open innovation on a high quality, robust software foundation. The increased profiling and performance capabilities, as well as the additional state-of-the-art model implementations will help researchers conduct broader AI explorations. Once a high-quality AI model has been developed, the next step is the clinical validation and evaluation process which MONAI Deploy addresses. Monai Fig 4.0: Training performance with MONAI v0.7 for 3D Spleen segmentation Figure 5.0: UNETR Network Architecture