MIMO: A Comprehensive Evaluation Method for Multi-Organ Medical Image Segmentation

Featuring encoder and decoder structures, U-Net and its variants have gained popularity in the field of medical image segmentation. However, research on model evaluation technology lags behind the model's development, with existing techniques facing challenges of complexity and uncertainty in guiding clinical practice and lacking comprehensiveness and unity. Aiming to quantitatively evaluate the performance of a model more comprehensively, we propose a method for evaluating models in multi-index and multi-organ segmentation tasks, dubbed as MIMO. MIMO simultaneously quantifies multi-organ segmentation results, multiple accuracy indices, and confidence estimates in a unified metric and provides concrete information on the overall performance of the model. Additionally, the size of the MIMO final fraction can be used to compare the model: A model with a larger score can be considered as more comprehensive and more available models. Experiments on eight different medical image segmentation models demonstrate that the proposed method offers novel insights and concise metrics for clinical model deployment.'}