Alvarez-Jimenez Charlems, Barrera Cristian, Munera Nicolas, Viswanath Satish E, Romero Eduardo
Annu Int Conf IEEE Eng Med Biol Soc. 2019 Jul;2019:2695-2698. doi: 10.1109/EMBC.2019.8856927.
Prostate cancer (PCa) diagnosis is established by pathological examination via biopsies, which are associated with significant complications and false negatives. Using MRIs to identify locations with high probability of containing cancer could instead be used to guide the biopsy procedure. The present investigation aims to identify target regions within different prostatic zones on MRI with high probability of being cancerous for assisting in the decision of where and how to perform biopsy. Our approach involved extracting multi-scale texture features for capturing local patterns to distinguish cancer and healthy tissue in different T2W-MRI prostate zones. Three different classification models were fed by the proposed strategy, namely support vector machine (SVM), Adaboost, and Random Forest. SVM with a linear kernel showed the best classification performance, with AUC scores of 0.91 in the anterior fibromuscular stroma area, 0.85 in the peripheral zone, and 0.87 when classification is performed independently of the prostate zone. The proposed method demonstrated that discriminant multi-scale texture features can accurately identify regions of prostate cancer in a zone-specific fashion, via MRI.
前列腺癌（PCa）的诊断通过活检的病理检查来确定，而活检会带来严重并发症和假阴性结果。使用磁共振成像（MRI）来识别极有可能含有癌细胞的位置，转而可用于指导活检程序。本研究旨在识别MRI上不同前列腺区域内极有可能癌变的目标区域，以协助决定在何处以及如何进行活检。我们的方法包括提取多尺度纹理特征以捕捉局部模式，从而区分不同T2加权磁共振成像（T2W-MRI）前列腺区域中的癌组织和健康组织。通过所提出的策略，为三种不同的分类模型提供数据，即支持向量机（SVM）、自适应增强（Adaboost）和随机森林。具有线性核的支持向量机表现出最佳的分类性能，在前纤维肌基质区域的曲线下面积（AUC）得分为0.91，在外周区为0.85，在独立于前列腺区域进行分类时为0.87。所提出的方法表明，判别性多尺度纹理特征能够通过MRI以区域特异性方式准确识别前列腺癌区域。
