基于自适应能量偏移场无边缘主动轮廓模型的乳腺肿块分割与分类方法研究

目的:为提高乳腺癌检测的精准度和效率，提出了一种基于自适应能量偏移场无边缘主动轮廓模型（AEOF-CV）的乳腺肿块分割与分类方法。方法:首先采用中值滤波、阈值分割及区域连通进行图像预处理，去除图像噪声；然后使用伽马变换及形态学运算相结合的方法进行图像增强；其次，采用AEOF-CV对弱对比度图像提高分割精度，用于乳腺肿块分割，得到感兴趣区域；最后使用不同提取特征方法，结合支持向量机识别感兴趣区域是否有肿块，并对存在肿块的图像判别肿块的良、恶性。结果:实验利用DDSM数据库中350个图像进行测试，实验结果证明，基于AEOF-CV乳腺肿块分割方法可以得到肿块清晰外部轮廓，具有较好的鲁棒性，误分率可达到0.212 0。无肿块样本识别率达到94.57%，恶性肿块识别率为97.91%，良性肿块识别率为96.96%，总识别率达94.00%。结论:基于AEOF-CV的乳腺肿块分割效果较好，误分率相对CV方法降低19.17%，查准率和查全率达到了0.851 9和0.836 5，全局分析性能较好，是乳腺肿块分割的有效方法，可为后续模式识别提供可靠依据。

Breast mass image segmentation and classification based on adaptive energy offset field-CV

Abstract: Objective To propose a method based on adaptive energy offset field-CV (AEOF-CV) for breast mass image segmentation and classification, thereby improving the accuracy and efficiency of breast cancer detection. Methods Firstly, median filtering, threshold segmentation and regional connectivity were used for image preprocessing to remove image noise. Then the image was enhanced by combining gamma transformation and morphological operation. Subsequently, AEOF-CV was used to improve the accuracy of low-contrast image segmentation for realizing breast mass image segmentation and obtaining the regions of interest. Finally, different feature extraction methods were combined with support vector machine for identifying whether there was a mass in the regions of interest and whether the mass was benign or malignant. Results A total of 350 images in DDSM database were tested. The experimental results showed that breast mass image segmentation based on AEOF-CV could obtain a clear external contour of the mass, with good robustness, and the misclassification rate was 0.212 0. The recognition rate for non-mass samples was 94.57%, and the recognition rates for malignant masses and benign masses were 97.91% and 96.96%, respectively. The average recognition rate of the proposed method reached 94.00%. Conclusion Breast mass image segmentation based on AEOF-CV has a good performance, with the misclassification rate reduced by 19.17% as compared with CV method, and the precision and recall rates are up to 0.851 9 and 0.836 5. The proposed method which has a good global analysis performance is an effective method for breast mass image segmentation and can provide a reliable basis for subsequent pattern recognition.Keywords: breast mass image segmentation energy offset field CV model support vector machine