查看更多>>摘要：News editors obtained the following quote from the background information suppli ed by the inventors: “Machine learning models are routinely employed in the fiel d of medical image processing and medical image analysis. In one example, machin e learning models may be used to segment a medical image into different anatomic al regions, or other regions of interest, by predicting/inferring one or more se gmentation masks for the regions of interest, wherein a segmentation mask identi fies pixels/voxels of an input image belonging to a particular region of interes t. One recognized limitation of the segmentation masks produced via machine lear ning models is the inability to distinguish between more and less confident segm entation masks, or between more and less confident regions within a segmentation mask. In conventional approaches, uncertainty of a machine learning model’s pre dictions is quantified during a training phase (e.g., accuracy, precision, recal l, cross-entropy etc.), and thus a same uncertainty may be associated with segme ntation masks produced by a particular machine learning model, regardless of the quality or identity of the input image. Therefore, in conventional approaches, it may be difficult for a user to determine if the information presented by a se gmentation mask, or derived therefrom, should be accepted or rejected. In partic ular, in anatomical measurement workflows, or other workflows where a position o f an anatomical landmark or other feature may be automatically determined based on output from a machine learning model (e.g., a segmentation mask), it may be d ifficult to accurately assess the certainty/confidence of the automatically dete rmined landmark position. In such cases, a radiologist or other user may need to manually evaluate each automatically determined position before proceeding, the reby reducing the efficiency of medical imaging workflows. Therefore, exploring approaches for automatically determining anatomical landmark position certainty/ confidence is generally desired.”
NETL
NSTL