Cardoso Jaime S, Cardoso Maria J
Faculdade de Engenharia and INESC Porto, Universidade do Porto, Campus da FEUP, Rua Dr. Roberto Frias, no. 378 4200-465 Porto, Portugal.
Artif Intell Med. 2007 Jun;40(2):115-26. doi: 10.1016/j.artmed.2007.02.007. Epub 2007 Apr 8.
This work presents a novel approach for the automated prediction of the aesthetic result of breast cancer conservative treatment (BCCT). Cosmetic assessment plays a major role in the study of BCCT. Objective assessment methods are being preferred to overcome the drawbacks of subjective evaluation.
The problem is addressed as a pattern recognition task. A dataset of images of patients was classified in four classes (excellent, good, fair, poor) by a panel of international experts, providing a gold standard classification. As possible types of objective features we considered those already identified by domain experts as relevant to the aesthetic evaluation of the surgical procedure, namely those assessing breast asymmetry, skin colour difference and scar visibility. A classifier based on support vector machines was developed from objective features extracted from the reference dataset.
A correct classification rate of about 70% was obtained when categorizing a set of unseen images into the aforementioned four classes. This accuracy is comparable with the result of the best evaluator from the panel of experts.
The results obtained are rather encouraging and the developed tool could be very helpful in assuring objective assessment of the aesthetic outcome of BCCT.
本研究提出了一种用于自动预测乳腺癌保乳治疗（BCCT）美学效果的新方法。美容评估在BCCT研究中起着重要作用。为克服主观评估的缺点，客观评估方法更受青睐。
将该问题作为模式识别任务来解决。一组国际专家将患者图像数据集分为四类（优秀、良好、中等、差），提供了金标准分类。作为可能的客观特征类型，我们考虑了领域专家已确定的与手术美学评估相关的特征，即评估乳房不对称、肤色差异和疤痕可见度的特征。基于从参考数据集中提取的客观特征开发了一种基于支持向量机的分类器。
将一组未见图像分类到上述四类时，获得了约70%的正确分类率。这一准确率与专家小组中最佳评估者的结果相当。
所获得的结果相当令人鼓舞，所开发的工具在确保对BCCT美学结果进行客观评估方面可能非常有帮助。
