蛛网膜下腔出血计算机辅助诊断的现状与展望

蛛网膜下腔出血由于通过蛛网膜下腔循环而导致出血信号难以检测,从而导致对该疾病的漏诊率可高达25%。本文综述了蛛网膜下腔出血计算机辅助诊断的现状:蛛网膜下腔出血在影像上可以表现为脑沟消失和/或对比度低的高信号,难以用传统的图像分割方法分割出血信号。现有的方法是先通过图谱配准或距离变换估计蛛网膜下腔,然后通过机器学习识别在估计的蛛网膜下腔是否有信号异常。现有算法的问题在于少量出血的高信号以及部分脑沟消失容易漏检。本文较为详细地介绍了基于距离变换估计蛛网膜下腔并基于支持向量机的蛛网膜下腔出血识别,为进一步提高诊断率,提出了可能的发展方向,即研究新的图像分析算法,实现低对比度高信号的检测,并准确地量化脑沟。

State-of-the-art and Perspective for Computer Assisted Diagnosis of Subarachnoid Hemorrhage

Misdiagnosis of subarachnoid hemorrhage (SAH) is as high as 25%, due to the difficulties discerning bleeding since SAH re-circulates within the subarachnoid space to make the change in grayscales of bleeding very subtle. For computer assisted diagnosis (CAD) of SAH, its state-of-the-art is reviewed: SAH can be in the form of effacement of sulci or high signal with low contrast on computed tomography (CT) images, and is difficult to be segmented using traditional segmentation methods; existing CAD system of SAH consisted of 2 steps (approximation of subarachnoid space via atlas registration or distance transformation and judging abnormalities of grayscale distribution in the approximated subarachnoid space by means of pattern recognition), and can yield erroneous conclusions when the bleeding is small or effacement of sulci occurs. Details of the algorithms are given to approximate subarachnoid space based on distance transform and to recognize SAH based on support vector machine. Possible ways to enhance the performance of CAD of SAH are pointed out: to develop new image processing methods such that high signals with low contrast as well as sulci can be well segmented and quantified.