Prostate cancer is one of the most common forms of cancer in men in many countries. The survival rate can be significantly enhanced with early detection of the cancer so that appropriate intervention can be administered. In this work, a novel automated classification algorithm by fusing a number of deep learning approaches has been proposed to detect prostate cancer from ultrasound (US) and MRI images. In addition, the proposed method explains why a specific decision is made given the input US or MRI image. Several pre-trained deep learning models having customs-developed layers are added on the top of the respective pre-trained models and applied to the datasets. The best model generates a maximum accuracy of 97% on US images and 80% on MRI images of the test set. The model that produced the best classification performance was selected to use as feature extractor from the dataset to build a fusion model as a next step. To improve the models performance, especially on the MRI dataset, a fusion model is developed by combining the best performing pre-trained model as feature extractor with some other shallow machine learning algorithms (e.g., SVM, Adaboost, K-NN, and Random Forests). This fusion approach remarkably improves the performance of the system by achieving the accuracy from aforementioned 80% to 88% on the MRI dataset. Finally, the fusion model is examined by the explainable AI to find the fact why it detects a sample as Benign or Malignant Stage in prostate cancer. The proposed approach can be adopted in smart clinics or hospitals for efficient prostate cancer detection and explanation.

在许多国家，前列腺癌是男性最常见的癌症形式之一。癌症的早期检测可以显着提高存活率，从而可以进行适当的干预。在这项工作中，提出了一种融合多种深度学习方法的新型自动分类算法，用于从超声 (US) 和 MRI 图像中检测前列腺癌。此外，所提出的方法解释了为什么在给定输入 US 或 MRI 图像的情况下做出特定决定。在各个预训练模型的顶部添加了几个具有自定义开发层的预训练深度学习模型，并应用于数据集。最佳模型在 US 图像上生成的最大准确率为 97%，在测试集的 MRI 图像上生成的最大准确率为 80%。选择产生最佳分类性能的模型用作数据集中的特征提取器，以构建融合模型作为下一步。为了提高模型性能，尤其是在 MRI 数据集上，通过将性能最佳的预训练模型作为特征提取器与其他一些浅层机器学习算法（例如 SVM、Adaboost、K-NN 和随机森林）相结合，开发了融合模型）。这种融合方法通过在 MRI 数据集上实现从上述 80% 到 88% 的准确度，显着提高了系统的性能。最后，融合模型由可解释的 AI 进行检查，以找出其将样本检测为前列腺癌良性或恶性阶段的原因。所提出的方法可用于智能诊所或医院，以进行高效的前列腺癌检测和解释。