区域生长法结合多竞争最小二乘拟合算法去除乳腺X线摄影图像中胸大肌影

目的 评价区域生长法结合多竞争最小二乘拟合算法去除数字乳腺X线摄影（MG）图像中胸大肌影的价值。方法 分层抽样法随机抽取244例MG数据，对图像进行轮廓选择、增强数据特征、胸大肌边界轮廓粗定位和去噪处理；结合最小二乘法改进区域生长法，拟合胸大肌的边界轮廓函数，使用最优轮廓函数制作胸大肌掩膜图，计算预测图与人工勾画图交并比（IOU）及像素精度（PA），评价其去除MG图像中的胸大肌影的价值。结果 基于上述方法所获胸大肌轮廓较为平滑，较少漏分割或过度分割，结果误差较小；还原胸大肌边界轮廓与手动分割结果非常接近，平均IOU为（89.76±4.28）%，平均PA为（89.98±3.91）%。结论 结合区域生长法与多竞争最小二乘拟合算法可用于去除MG图像中的胸大肌影。

Regional growth method combining with multi-competitive least-squares for removal of pectoralis major in mammographic images

Objective To observe the value of regional growth method combining with multi-competitive least-squares fitting algorithm to remove the pectoralis major shadow from images of digital mammography (MG). Methods MG data of 244 cases were randomly selected using stratified sampling method. Contour selection, enhanced data features, coarse localization of pectoralis major boundary contour and denoising processes were performed on the images. Combining with the least-squares method, improved region growth method was used to fit the boundary contour function of the pectoralis major, while the optimal contour function was used to produce the pectoralis major mask images. Intersection over union (IOU) and pixel accuracy (PA) between the predicted and manually outlined maps were calculated to evaluate the value of removing pectoralis major shadow on MG images. Results The pectoralis major contours obtained based on the above way were rather smooth, with fewer missed segmentation, over-segmentation and less error in the results, and the restored pectoralis major boundary contours were very close to those of manual segmentation, with the average IOU of (89.76±4.28)% and the average PA of (89.98±3.91)%. Conclusion Combining region growth method of multi-competitive least-squares fitting algorithm could be used to remove the pectoralis major shadow on MG images.