基于证据深度神经网络的医学影像三支决策

Three-way decisions for medical imageswith evidential deep neural networks

在医学领域中,基于影像数据分析的决策支持系统通常采用深度神经网络对医学影像进行分类以实现智能辅助诊断。然而,受标注数据受限、数据噪声、病灶表征不明确等因素影响,实际应用中医学影像数据通常包含不确定样例。对不确定影像数据直接应用深度学习方法进行强制确定分类可能导致严重病例误判,从而造成过高的系统决策风险。作为不确定性决策的有效工具,三支决策为不确定性数据分类提供了理论依据。基于三支决策的病例数据分类有助于区分不确定病例,延迟谨慎决策,通过人机协同有效降低系统决策风险。传统三支决策理论方法基于逻辑规则构造,擅长分析结构化数据,对于图像等非结构化数据不确定性分类存在应用局限。针对该问题,文中将证据理论与深度神经网络相结合,拓展构造了证据深度神经网络,并针对医学影像分类任务,构造了基于证据深度神经网络的医学影像三支决策方法。在肺炎、乳腺癌影像数据集上的实验表明,提出的三支决策方法能够有效识别高不确定度医学影像,降低决策系统病例分类风险。

In medical decision support systems,deep neural networks are widely utilized to classify medical images to realize the intelligent diagnosis.Due to the limitation of training data,data noise and unclear disease symptoms,the medical images in real applications usually contain uncertain samples.Applying certain classification methods to uncertain medical images for farfetched classification may lead to serious diagnosis mistakes and result in high decision risks.As an effective tool for uncertain decision making,Three-way decisions provides a theoretical basis for the classification of uncertain data.The uncertain classification of medical images based on Three-way decisions is helpful to distinguish uncertain cases for delayed decision making and the cautious diagnosis through man-machine cooperation.However,the traditional three-way decision methodology is not effective to handle digital images.To tackle the problem,we combine the evidence theory with deep neural networks to construct evidential deep neural networks and thereby implement a three-way decision method for uncertain classification of medical images.Experiments on pneumonia and breast cancer image data sets validate that the proposed three-way decision method is effective to identify medical images with high uncertainty and reduce the risk of medical image classification in decision support systems.