前列腺癌是男性第二大致死癌症，近年来我国的前列腺癌发病率呈明显上升趋势，在所有癌症中排名第五。基于多参数核磁影像分析的前列腺癌自动检测诊断是提高前列腺患者存活率、改善预后效果的关键，成为当前国内外研究的热点前沿课题。本项目拟结合机器学习方法和思想，重点研究面向小数据量训练样本情况下的病灶区域检测、分割、特征学习以及样本分布严重不均衡情况下的分类问题，具体开展以下几个方面的研究：（1）多参数核磁影像的信号归一化方法；（2）面向精确分割的端到端图像配准方法；（3）小训练样本情况下的精确目标检测和分割方法；（4）多角度特征学习和融合方法；（5）样本分布严重不均衡情况下的特征学习和分类方法。最后，拟搭建前列腺癌自动检测诊断示范系统，对项目研究的方法进行测试和评价。本项目为大规模前列腺癌筛查、活检穿刺辅助导航以及其他典型疾病的医学影像辅助诊断等实际应用提供关键的核心算法和技术支撑。

Prostate cancer (PCa) is the second leading cause of death among men. Recently, the incidence of PCa in China is on the rise, ranked 5 among all cancers. Automated detection and diagnosis of PCa based on multiparametric magnetic resonance images (mp-MRI), which is critical for increasing the survival rate of patients and improving prognosis, has been become a world-wide hot research topic. In the proposal, we will integrate the methods and concept of machine learning into the mp-MRI based PCa detection and diagnosis task. Our research mainly focuses on accurate foci region detection, segmentation and feature learning based on only a very limited amount of training data, as well as image classification based on severely unevenly distributed samples. More specifically, we will conduct in-depth exploration from the following five aspects: (1) mp-MRI data normalization methods, (2) accurate segmentation oriented end-to-end image registration method, (3) accurate object detection and segmentation based on limited training data, (4) multi-view feature learning and fusion, and (5) feature learning and classification based on severely unevenly distributed samples. Finally, we will build an automated PCa detection and diagnosis system to evaluate the proposed methods. This proposal can provide some theoretical and technical support for large-scale PCa screening, navigating prostate biopsies and computer-aided diagnosis of many other typical diseases from medical images.