With the development of smart justice in China, the Supreme People’s Court implements the system of compulsory retrieval for similar cases continuously and comprehensively, more and more judicial big data has been continuously disclosed, and the application of case retrieval is more extensive, and then, the accuracy of similar case search results needs to be urgently improved. Legal case retrieval is a special search task; for a given query case, it refers to the searching for similar cases. Different from traditional text search, legal case retrieval has different characteristics and greater challenges, for its query case is longer and more complex than common keyword queries and short article queries. In addition, the definition of dependencies between query cases and candidate cases differs from general dependencies based on text or topic. In order to solve these problems, we propose a method for similarity case retrieval based on the legal facts, and our model combine the topic distribution and legal entity facts to make the document representation vector more suitable for legal scenarios. At the same time, the method of paragraph aggregation based on BERT is used to encode context semantic information and solve the problem of long text. The experimental results show that our method is superior to the existing method.

Excellent performance has been achieved on medical image segmentation. Still, existing algorithms perform relatively poorly for annular objects with high intra-class variability and inter-class similarity, which easily leads to regional confusion, especially in medical images with annular regions. In this paper, we first propose an annular prior prompt learning method based on MedSAM, coined APPL, which not only can strengthen the learning for annular region features, but also effectively handles the regional confusion caused by high intra-class variability and inter-class similarity. Specifically, a novel annular prior prompt encoder (APPE) is proposed based on our designed annular constraint to alleviate the negative impact of inter-class similarity, which maps multiple prompt points into a linear regression feature space to provide the standardized annular prompt feature for the mask decoder. In addition, a new region connectivity enhanced image encoder (RCEIE) is developed to introduce morphology attention into supervised training, which reduces feature noise caused by intra-class variability, significantly improving the connectivity of region features. Powered by the collaboration across different annular features, the mask decoder can effectively avoid inter-class confusion to reduce the impact of weak boundaries, further boosting the performance for tackling highly diverse pixel distributions in medical images. Compared with most state-of-the-art methods, the proposed method achieves 90.07% of the mIoU and 94.71% of the DSC on an in-house dataset, and shows strong generalization on two public heart segmentation datasets with end-diastolic DSC scores of 94.2% and 83.9%, and end-systolic DSC scores of 95.5% and 87.3%. The experimental results demonstrate the consistently superior performance of our method quantitatively and qualitatively.