基于SIFT特征关键点匹配的脑组织提取方法

脑组织提取是脑部磁共振图像处理的重要步骤之一。为稳定准确地提取脑组织,提出一种基于点阵SIFT(Scale Invariant Featun Transform)特征匹配的脑组织提取方法。该方法在采用BET方法对脑组织进行提取的过程中,不断提取轮廓点阵的SIFT特征,并提出基于距离约束投票的特征匹配方法,对目标图像当前脑组织轮廓和模板图像脑组织进行匹配,根据匹配结果重新定义BET脑组织提取参数再继续进行轮廓演化,从而得到稳定的脑组织轮廓。然后以该轮廓为初始值采用Graph cuts方法进行精提取,最终得到精确的脑组织边界。采用IBSR网站20套MRI图像序列与其他3种算法进行对比实验,得到最高的DICE精度(0.962±0.008)、Jaccard精度(0.926±0.014)和特异性(0.994±0.004)、最低的假阳性率(4.95%±2.74%)、较低的假阴性率(2.82%±2.0%)。精度最高表明该方法具有很好的提取精度,相应参数的标准差较小表明本方法有很好的稳定性,说明该算法是一种精确稳定的脑组织自动提取方法。

Brain Extraction Method Based on SIFT Feature Matching among Key Points

Brain extraction is an important preprocessing step in most of image analyses for cerebral MRI image. ASIFT(Scale Invariant Featun Transform) feature matching based brain extraction method was proposed in this paper for the precise and robust brain extraction. This method combined the tessellated surface based method (BET) and atlas based method. Firstly,the BET method was used to evolve the brain contour, and the SIFT features of the key points on brain contour were extracted iteratively to find the matched key points between the target image and atlas image using weak constraint of spatial distance based matching method. Next, the matched points were used to renew the parameter of BET and the brain contour iteratively, thus a robust brain contour was obtained. At last, a graph cuts method was used to refine the brain contour to obtain a precise brain contour. Twenty MRI volumes from IBSR web site were tested using this method and other three methods, and the proposed method obtained the highest Dice precision (0.962±0.008), Jaccard precision (0.926±0.014) and specificity (0.994±0.004), as well as the lowest false positives rate (4.95%±2.74%) and lower false negatives rate (2.82%±2.0%). The highest extraction precisions demonstrated this method precise, and the lower corresponding standard deviations showed this method was robust. Therefore, this proposed method is a precise and robust brain extraction method.