Automatic segmentation of meningioma from non-contrasted brain MRI integrating fuzzy clustering and region growing

Abstract

Background: In recent years, magnetic resonance imaging (MRI) has become important in brain tumor diagnosis. Using this modality, physicians can locate specific pathologies by analyzing differences in tissue character presented in different types of MR images.This paper uses an algorithm integrating fuzzy-c-mean (FCM) and region growing techniques for automated tumor image segmentation from patients with menigioma. Only non-contrasted T1 and T2 -weighted MR images are included in the analysis. The study's aims are to correctly locate tumors in the images, and to detect those situated in the midline position of the brain.

Methods: The study used non-contrasted T1- and T2-weighted MR images from 29 patients with menigioma. After FCM clustering, 32 groups of images from each patient group were put through the region-growing procedure for pixels aggregation. Later, using knowledge-based information, the system selected tumor-containing images from these groups and merged them into one tumor image. An alternative semi-supervised method was added at this stage for comparison with the automatic method. Finally, the tumor image was optimized by a morphology operator. Results from automatic segmentation were compared to the "ground truth" (GT) on a pixel level. Overall data were then evaluated using a quantified system.

Results: The quantified parameters, including the "percent match" (PM) and "correlation ratio" (CR), suggested a high match between GT and the present study's system, as well as a fair level of correspondence. The results were compatible with those from other related studies. The system successfully detected all of the tumors situated at the midline of brain.Six cases failed in the automatic group. One also failed in the semi-supervised alternative. The remaining five cases presented noticeable edema inside the brain. In the 23 successful cases, the PM and CR values in the two groups were highly related.

Conclusions: Results indicated that, even when using only two sets of non-contrasted MR images, the system is a reliable and efficient method of brain-tumor detection. With further development the system demonstrates high potential for practical clinical use.

Figure 1

Study material. One of the 29 study groups of the MR images of menigioma is shown here, T1-wighted image (left), T2-weighted image (middle) and two-dimensional intensity histogram based on these 2 images (right).

Figure 2

Flow Chart of the research procedures.

Figure 3

Result after FCM clustering. FCM clustering on MR images of the brain (left), and the histogram after defuzzification (right), a total of 32 groups of binary image were produced according to the colors zones shown on the histogram, each color represent a particular tissue character.

Figure 4

Result after Region growing. After FCM clustering, some image may be too fragmenting to be properly classified (upper left and lower left). In this occasion, a seed area is selected within the image (upper middle and lower middle), and after pixel aggregation, these fragment could grow into more meaningful image. The upper row image now could be identified as tumor-containing image, where as the lower row image will be classified as background, and be readily processed at later stage.

Figure 5

The bounding box method. Using brain tissue image (left 1) as standard, the image group whose image box is larger than 1/2 height and width of the brain image will be excluded (left 2), otherwise will be preserved (right 1 and 2).

Figure 6

The difference of bilateral side histogram analysis. By analyzing the T2-weighted MR image (left), we obtained the histogram of bilateral brain tissue (right upper) and the curve represented their difference (right lower). Using the highest absolute value in histogram difference curve as reference point (red line), here the grayscale value of both sides histogram were compared, and the side with higher grayscale value is likely to have tumor. In this case, the left (green) side had higher grayscale at reference point, so the tumor is on the left side of the brain.

Figure 7

The solidity of tumor image. Left: image with high solidity (78.99) more likely to be tumor; Right: image with low solidity (39.77) more likely to be normal tissue.

Figure 8

The result of tumor segmentation. One of the result were shown here, the original non-contrasted T1 (upper left) and T2 -weighted (upper middle) MR image were processed. The tumor image segmented by semi-supervised method (lower left) and automatic method (lower middle) were compared with "ground truth" (lower right), which was manually segmented from contrasted-enhanced T1 image (upper right).

Figure 9

Diagram showing quantified evaluation of the results obtained by automatic pathway and semi-supervised alternative. Upper: percent match (PM) curves of 29 cases; Lower: correspondence ratio (CR) curves of 29 cases in two groups. We can see in case no. 2, 11, 15, 17 and 18, low PM and CR were observed in automate pathway but not in semi-supervised group; Case 29 show poor PM but good CR in both groups. Beside these cases, some trace difference could be observed in case no. 22, 26 and 28.

Figure 10

The relation between cluster number in FCM and tumor image separation rate. Could see the separation rate is improving as cluster number increase; finally come to a plateau when cluster numbers exceed the number of 32.