基于局部上下文引导特征深度融合的轻量级医学图像分割方法OA

For most of the existing deep learning-based medical image segmentation,a large amount of training data were used to improve the detection network to achieve excellent detection performance.These methods needed a large number of the models and parameter,resulting in poor real-time detection performance.To address this,a local context guided feature deep fusion lightweight medical segmentation network(LCGML-net)was proposed.The main idea of LCGML-net was to reduce the number of parame-ters required for model fitting by accurate feature selection and feature fusion,thus achieving smaller model while maintaining detection accuracy.LCGML-net enriched the feature representation accurate pre-cision segmentation of the target by extracting dense multi-level and multi-scale local context features of the target in the feature extraction stage and the feature mapping stage,respectively.Extensive experi-ments were conducted on multiple medical segmentation benchmark datasets,including STARE,CHASEDB1,and KITS19.The results demonstrated that compared to other advanced methods,the pro-posed LCGML-net exhibited the best detection performance with the smallest model parameters.