基于格式塔认知框架的乳腺肿块分割算法

Breast mass segmentation algorithm based on gestalt psychology framework

针对X线图像乳腺肿块分割易受边缘及周围腺体组织干扰,分割精度不高的问题,该文提出了一种基于格式塔认知框架的乳腺肿块分割算法。该算法利用格式塔心理学理论,对人类视觉自下而上的感知和自上而下的认知过程建模,并将其在肿块分割问题中实例化表示。首先,抽取视觉块,并将其作为基本认知单元;然后,利用图像局部自相似性及格式塔规则进一步简化图像;最后,从全局特征出发,融入专家诊断知识,通过最优化实现肿块的自动化分割。在公开数据集INbreast上进行实验,对比其他流行算法,分割准确率提高了10%。该算法实现了无监督的自动化病灶分割,无需人工干预,对图像噪声具有强抗干扰性。

The CAD system for breast mass segmentation is a challenging task due to the abundant glandular tissue of X-ray image.This paper proposes a novel breast mass segmentation method based on gestalt psychology framework.The framework explores the cognitive process of human including bottom-up sensation and top-down recognition.Then,it is instantiated in the application of breast mass segmentation problem.Firstly,the visual patches are generated as basic processing units.Secondly,the visual patches are further simplified by the self-similarity of image and gestalt psychology.Finally,prior experience of radiologists is integrated into the framework.The final segmentation results are obtained by a global optimization method.The experiments are implemented on the public INbreast dataset and the segmentation accuracy increases 10%comparing with other popular algorithms.It concludes that the proposed method can be used to achieve automated segmentation of mass legions.Meanwhile,it yields stronger robustness.