基于磁共振影像的乳腺癌计算机辅助检测方法综述

Review of breast cancer computer-aided detection methods based on magnetic resonance imaging

乳腺癌已超越肺癌,成为全球第一大癌症,其发病率和死亡率在女性中均为首位。目前,磁共振成像在乳腺癌诊断和评估方面已得到广泛应用,是检测敏感度最高的成像方法。计算机辅助检测系统对辅助医生诊断乳腺癌具有重要作用,其中,准确的乳腺癌检测是关键技术之一。综述近年来基于磁共振成像的乳腺癌检测方法,旨在为深入开展乳腺癌检测研究提供参考。首先,对乳腺癌计算机辅助检测的研究意义及检测流程进行介绍;然后,对乳腺癌检测的传统方法和深度学习方法的优缺点进行分析;最后,对乳腺癌检测技术的现有问题和未来发展方向进行总结与展望。传统乳腺癌检测方法的结果可解释性强但泛化能力差,暂时无法满足临床需求。近年来深度学习检测方法飞速发展,检测性能大幅提高,因此着重介绍深度学习检测方法在乳腺磁共振影像中的最新应用。基于磁共振影像的乳腺癌计算机辅助检测方法已取得了较大的成就,但肿块型和非肿块型乳腺癌的特征差异给检测带来了困难,高质量数据库的紧缺仍是深度学习方法的瓶颈。因此,多模态影像的应用和弱监督深度学习方法将会是未来研究者重点关注的方向。

Breast cancer has overtaken lung cancer as the world’s most common cancer. Its incidence rate and mortality rate are the highest among women. At present, magnetic resonance imaging has been widely used in breast cancer diagnosis and evaluation, and is the imaging method with the highest detection sensitivity. Computer aided detection system plays an important role in assisting doctors to diagnose breast cancer. Accurate breast cancer detection is one of the key technologies. In order to provide reference for further research on breast cancer detection, the breast cancer detection methods based on magnetic resonance imaging in recent years were reviewed. Firstly, the research significance and detection process of breast cancer computer-aided detection were introduced;then the advantages and disadvantages of traditional and deep learning methods for breast cancer detection were analyzed;finally, the existing problems and future development direction of breast cancer detection technology were summarized and prospected. The results of traditional breast cancer detection methods are interpretable but poorly generalizable, and they temporarily cannot meet clinical needs. In recent years,the deep learning detection method has developed rapidly, and the detection performance has been greatly improved. Therefore, the latest application of deep learning detection method in breast magnetic resonance imaging is emphasized. Breast cancer computer-aided detection methods based on magnetic resonance imaging have made great achievements, but the difference in characteristics between mass and non-mass type breast cancer causes difficulties in detection. The shortage of high-quality databases remains a bottleneck in deep learning methods. Therefore, the application of multimodal images and weakly supervised deep learning will be the focus of future researchers.