Medical & Biological Engineering & Computing ( IF 2.6 ) Pub Date : 2021-04-05 , DOI: 10.1007/s11517-021-02348-4 Sunita Sarangi 1 , Nrusingha Prasad Rath 2 , Harish Kumar Sahoo 2
|
Breast cancer is a leading cause of mortality affecting women across the world. Early detection and diagnosis can decrease the mortality rate due to this cancer. Machine learning-based models are gaining popularity for biomedical applications due to the ability of nonlinear mapping between input and output patterns using supervised training phase. The research work in the paper is focused on the optimal adaptive threshold for mammogram mass segmentation, and detection in order to assist radiologist in accurate diagnosis Legendre neural network with single layer is used to develop the model, and the training is performed through Block Based Normalized Sign–Sign Least Mean Square (BBNSSLMS) algorithm. Legendre neural network expands the input vector using standard Legendre polynomial, and the recursive update principle is followed for the weight vector in higher dimension. The optimal threshold is indirectly used for proper segmentation of mammogram mass. The proposed segmentation method involves training phase with 30 images and testing phase by 151 images obtained from standard Mammogram Image Analysis Society (MIAS) database. The proposed model achieved a sensitivity of 95% and accuracy of 96% with false positives per image calculated as 1.19.
Graphical abstract
中文翻译：
使用基于勒让德神经网络的最优阈值进行乳房X线图像质量分割和检测
乳腺癌是影响世界各地女性的主要死亡原因。早期发现和诊断可以降低这种癌症的死亡率。由于使用监督训练阶段在输入和输出模式之间进行非线性映射的能力，基于机器学习的模型在生物医学应用中越来越受欢迎。论文的研究工作集中在乳房X线图像质量分割的最佳自适应阈值，检测以辅助放射科医师准确诊断使用单层Legendre神经网络开发模型，并通过Block Based Normalized进行训练符号-符号最小均方 (BBNSSLMS) 算法。Legendre 神经网络使用标准 Legendre 多项式扩展输入向量，高维权重向量遵循递归更新原则。最佳阈值间接用于适当分割乳房 X 光照片质量。所提出的分割方法涉及使用 30 张图像的训练阶段和使用从标准乳房 X 光检查图像分析协会 (MIAS) 数据库获得的 151 张图像的测试阶段。所提出的模型实现了 95% 的灵敏度和 96% 的准确度，每张图像的误报计算为 1.19。
京公网安备 11010802027423号