Purpose To develop an accurate, effective melanoma computer-aided diagnosis algorithms in vivo based on laser scanning confocal microscope images. Materials and Methods Forty patients with melanoma and forty patients with benign nevus were performed with laser scanning confocal microscope imaging, and the texture features were extracted using wavelet analysis. The standard deviation, energy and entropy characteristic parameter were applied with classification and regression tree to automatic classification based on wavelet coefficient. Results The algorithm showed that 95.00%of melanoma and 92.50% of benign nevus could be correctly classified. Conclusion This computer-aided diagnosis algorithms can improve the accuracy of melanoma diagnosis, and also can decrease the misdiagnostic rate of benign nevus, which provides an objective evidence for the early detection and diagnosis of melanoma.% 目的基于激光共聚焦扫描显微镜皮肤图像，研发一种能够准确、有效地识别在体黑色素瘤的计算机辅助医学诊断方法.资料与方法通过小波分析法，提取40例黑色素瘤和40例常见良性痣患者激光共聚焦扫描显微镜图像的纹理特征，基于小波系数的标准差、能量以及熵值特征参数，采用分类与回归树算法对图像进行自动分类.结果该算法对对良性痣正确分类率达92.50%.结论该计算机辅助诊断方法不但提高了恶性黑色素瘤早期诊断的准确度，还降低了良性痣的误诊率，为临床早期发现和诊断黑色素瘤提供了客观依据.
展开▼
