华中科技大学学报（自然科学版）2025，Vol.53Issue(5)：135-142,8.DOI:10.13245/j.hust.250198
基于多分支结构和多层级注意力的乳腺肿块检测
Breast mass detection based on multi branch structure and multi-level attention
摘要
Abstract
Aiming at the problems of high missed detection rate and misdiagnose in existing breast mass detection algorithms for irregular lumps,a mass detection network model based on multi branch structure and multi-level attention was proposed.To solve the problem of poor expression ability of small tumor features,a multi branch structure consisting of asymmetric blocks and dilated convolution was designed to fully extract and fuse multi-level features while reducing the number of parameters,and an attention mechanism based on multi-level feature maps was designed,in which the attention module fused deep attention information and shallow feature map information between neighboring layers to obtain important features and their locations.To alleviate the issue of imbalanced positive and negative samples for the mass lesions,an adaptive sample partitioning algorithm was introduced in the sample selection process,which was based on the distribution characteristics of the intersection over union(IOU)values set of anchor frames and masses,and then adaptively calculated the partitioning threshold to ensure the positive sample ratio and the effectiveness of training.Experiments were conducted on CBIS-DDSM datasets.Results show that the average accuracy of the network model reaches 86.10%,and the miss rate reaches 10.19%,which is better than that of other main-stream algorithms.关键词
目标检测/乳腺钼靶诊断/多分支结构/注意力机制/自适应样本划分Key words
target detection/mammography diagnosis/multi branch structure/attention mechanism/adaptive sample division分类
信息技术与安全科学引用本文复制引用
谈玲,梁颖,夏景明..基于多分支结构和多层级注意力的乳腺肿块检测[J].华中科技大学学报（自然科学版）,2025,53(5):135-142,8.基金项目
国家重点研发计划科技创新2030——"新一代人工智能"重大项目(2021ZD0112201) （2021ZD0112201）
江苏省产学研基金资助项目(BY2022459). （BY2022459）
