Keywords

1 Introduction

Biomedical imaging has developed rapidly in recent years. Recent researches and their experiments really helped in advancing diagnostic tools for medical field [1]. The various types of biomedical imaging technologies are magnetic resonance imaging, computed tomography scan, ultrasound, SPECT, PET and X-ray. Brain tumor is the most common disease in medical science. Detection of brain tumor in early stages can improve the prevention mechanism to higher level. The MRI technique is the most effective technique for brain tumor detection. MRI is better than X-rays which results in high-quality images [2].

2 Existing Methodology

Image fusion is a method of joining complementary information and multi-modality images of the same patient into an image. Therefore, the obtained image consists of more informative than the individual images alone. In feature level fusion, source images are partitioned into regions and features like intensities of pixel, edges or texture are used for fusion technique [3].

Feature level fusion between images is a challenging problem of inter-image variability such as pixel mismatches (scale, rotations and shifts), missing pixels, image noise, resolution and contrast [4]. The inaccuracies in feature representation can lead to poor fusion performance and lesser robustness of the feature representation. In addition, this also means that wrong feature representation can lead to wrong conclusion that reduces the reliability of medical image analysis in clinical settings [5].

Region-based image fusion of feature level would be highly efficient when compared to the pixel-based fusion methods. The fusion method has multi-modal images which are partitioned into regions using automatic segmentation process [3], and the image fusion is performed based on the rules of region-based fusion. The major disadvantage of the existing system is that fusion system passes information within each decomposition level so that the source image details are preserved expressing the artifacts [6, 7].

3 Proposed Methodology

The proposed system has several advantages that overcome the disadvantages of the existing feature level fusion system.

  1. 1

    Sensitivity to noise and blurring effects can be achieved.

  2. 2

    The use of SVM classifier aids in detecting the tumor in the early stage.

  3. 3

    The proposed method uses the kernel trick, so we can build in expert knowledge about the problem (Fig. 1).

    Fig. 1
    figure 1

    Block diagram of the proposed system

    Full size image

In this work, support vector machines (SVM) are supervised learning models that analyze data used in classification and regression analysis. In addition to performing linear classification, SVMs can efficiently perform a nonlinear classification using the kernel trick, by mapping their inputs into high-dimensional feature spaces [8]. An efficient classification method is proposed to recognize normal as well as abnormal MRI brain images.

SVM classifier is implemented to segment the affected portion of cancer. To segment the portion, first, we have to filter out the noise in the acquired image based upon the masking methodology [9]. The morphological function will be applied throughout the filtered image. Using morphological bounding, box will be drawn over the detected portion. Hence, by means of SVM classifier, the region enclosed by bounding box will be spitted out separately (Fig. 2).

Fig. 2
figure 2

Workflow of the proposed system

Full size image

The input images are scanned using MRI and will be stored in MATLAB. These stored images are displayed as a grayscale image of size 256*256. The colored images are converted into grayscale image using RGB to gray conversion. The obtained image may contain noise. White Gaussian noise is the most commonly occurring noise in the MRI images. Hence, the noise removal is mandatory for the tumor detection from magnetic resonance images [3]. There are many types of filters used for image noise removal. The images are preprocessed to filter out the noise from a grayscale converted image. Edge detection is the most vital part in tumor detection. A Gabor filter is a linear filter which is used for edge detection [6, 10].

The next process will be the morphological operation where the boundary of the tumor part is approximately sized out in red. During this process, adding and removing of pixels take place through which the border of the tumor is detected. During the process of segmentation, two output portions are appeared, namely the tumor area is indicated as green boundary and the second image shows the segmented tumor portion. After detecting and segmenting the tumor area, a message is forwarded to the doctor. Also, a message is sent to the guardian’s mobile phone through global system for mobile communication (GSM) which points out the corresponding stage of the tumor and suggests the respective medicines to be taken [11].

4 Results and Discussion

In this paper, the support vector machine is applied to segment the detected portion of cancer. To segment the portion, first, we have to filter the acquired image based upon the masking methodology. The morphological function will be applied and extracted throughout the filtered image. With the help of morphological bounding, the box will be drawn over the detected portion. Then, the region enclosed by bounding box will be taken out separately by means of SVM classifier.

Copy the collected images for the initial process to the current folder and then execute it. If any noise occurs, it will be completely cleared. Get the input image from the collected data and apply it for preprocessing. The resultant image will be preprocessed image, i.e., input image and filtered image. The input MRI image is obtained from https://www.insight-journal.org/midas/gallery for research work and simulation purpose (Figs. 3 and 4).

Fig. 3
figure 3

Window displaying input image [12]

Full size image
Fig. 4
figure 4

Window displaying preprocessed image

Full size image

Perform the feature extraction and obtain the corresponding feature extracted values corresponding to correlation, energy, contrast and homogeneity from the preprocessed image (Fig. 5).

Fig. 5
figure 5

Window with feature extraction values

Full size image

The next process will be the morphological operation where the boundary of the tumor part is approximately sized out in red. During this process, adding and removing of pixels take place through the border of the detected tumor (Fig. 6).

Fig. 6
figure 6

Image after morphological operation

Full size image

The image after morphological operation is appeared on the screen in which unwanted dots and lines are removed (Fig. 7).

Fig. 7
figure 7

Window showing classified image

Full size image

During the process of segmentation, two output portions are appeared, namely the tumor area which is indicated as green boundary and the second image shows the segmented tumor portion (Fig. 8).

Fig. 8
figure 8

Display of segmented tumor area

Full size image

After the process of segmenting the tumor area, it is being mailed to the doctor. The stage of the tumor (early or advanced) is indicated on the screen (Figs. 9 and 10).

Fig. 9
figure 9

Screenshot of sender’s mail page with resultant image

Full size image
Fig. 10
figure 10

Window displaying tumor stage

Full size image

Following the mail notification, a message is also sent to the guardian’s mobile phone which points out the corresponding stage of the tumor and suggests the respective medicines to taken (Fig. 11).

Fig. 11
figure 11

Screenshot of receiver’s message

Full size image

5 Conclusion

Initial stage of brain tumor is a major challenge in the medical field. Brain tumor detection is a tedious job because of the complex structure of brain. MRI images provide an easier method to detect the tumor and also to perform the surgical approach for its removal. The existing image fusion feature level technique has limitation in accuracy, exactness and ability to detect the tumor earlier. To overcome the existing system limitations, SVM classifier is used for early detection of tumor and stage classification. This method is comparatively best when referring to other available algorithms.