基于视觉认知特性的乳腺X线图像分析与理解

项目名称： 基于视觉认知特性的乳腺X线图像分析与理解

项目编号： No.61201294

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 电子学与信息系统

项目作者： 王颖

作者单位： 西安电子科技大学

项目金额： 25万元

中文摘要： 乳腺癌是女性癌症发病率和死亡率均居第二位的癌症，早期诊断和治疗是挽救患者生命的最有效手段。乳腺X线摄影是目前乳腺癌普查的首选方法，但早期隐匿性乳腺癌通常影像学表现不明显，且与正常组织相似，极易引起漏诊和误诊，给患者造成了极大隐患。为此，本课题拟研究乳腺图像的视觉认知特性，结合医生诊断过程中的认知行为和经验，以视觉认知计算、机器学习等领域的新方法为基础：1）利用多尺度几何分析模拟人眼视觉响应特性，研究高灵敏度的感兴趣区域定位方法；2) 提取感兴趣区域的语义特征，通过稀疏编码方法逼近视觉系统的稀疏和认知特性，建立视觉语义特征库；3) 研究能够有效融合医生认知经验的半监督认知分类方法；4) 深入分析医生诊断时的认知心理行为，建立感兴趣区域性质的风险评估分析模型。该项目研究成果为乳腺X线图像的分析和理解提供了新的研究思路，同时也为辅助医生有效提高临床诊断准确性提供了坚实的理论基础和技术保障。

中文关键词： 视觉特性；乳腺X线图像；哈希编码；直推式学习；隐语义模型

英文摘要： Breast cancer is rank the second place of cancer incidence and also mortality among women, early diagnosis and treatment is the most effective ways to save lives. Mammography is the most preferred method of breast cancer screening at present. However, the ambiguous characteristics of the early cancers and its similarity to normal tissues will all probeblely induce the error and miss on diagnosing, which will cause great risk to patients. To this end, the project intends to study visual cognitive characteristics of mammogram. The project will combine cognitive behavior and experience of doctors during diagnosis process with novel methods of visual cognition computing and machine learning, to develop: 1) high sensitivity location methods for region of interest, based on simulating the human visual response characteristics using Multiscale Geometric Analysis; 2) establish visual semantic feature library through extracting the semantic features of the region of interest and sparse coding method to approximate the sparse and cognitive characteristics of visual system; 3) study the semi-supervised cognitive classification method which combines with the cognitive experience of doctors; 4) through analyzing the cognitive psychology behavior of doctor during his diagnosed procedure, construct the risk assessment analysis

英文关键词： visual characteristic；mammography；hashing coding；transductive learning；latent semantic model