Article Preview
TopIntroduction
Cancer is a dangerous disease. It produces unpredictable cells, and also reproduces new cells and new cells break existing organs and cause the tumour to spread, thus harming the body. Tumor in the brain is classified as primary tumor and metastasis tumor, Primary tumor is one type of cancer which grows in the brain tissue whereas metastasis tumor affects the other organs in the human body. Primary brain tumor is further classified as benign and malignant In benign tumor structure is uniform without active cells and does not affect the nearby tissues whereas malignant tumor contains heterogeneous structure with active cells affects the other parts of the body. To predict the tumor growth in advance and reduce the death rate due to breast cancer for women’s an automated frame work is necessary. The common types of cancers that develop in breasts are ductal carcinoma in situ, invasive ductal carcinoma, inflammatory breast cancer, and metastatic breast cancer and our proposed work focus on invasive ductal carcinoma breast cancer images. For diagnostic and treatment purposes, medical professionals (radiologists and doctors) need to understand the type of tumors they are dealing with.
In this context, MRI (Magnetic Resonance Imaging) scanning is used to get the high contrast resolution images of tumors. Using MRI scanner, images with various dimensions and various angles such as T1-Axial, T2-Axial, T1-Weighted, T1 Contrast enhanced and post processed, are captured efficiently. The MRI images greatly help medical professionals study the details of various soft tissues and structures associated with the tumors in question and help them classify tumors. In the whole process, the technique of segmenting images of tumors from the MRI images plays an important role. MRI image sequences usually have non-uniform intensity variations and noise inference signals. They prevent enhanced visualisation of images and proper tumor segmentation. Image segmentation partitions images into small number of homogenous regions based on their size, texture and pixel intensity. Thus it enhances the vision of the tumor region. This helps medical professional to extract the meaningful information from scanned images. From the clinical prospective, when segmentation is done manually, it takes a lot of time. Further manual segmentation can result in human errors. To overcome the above said issues, an automated segmentation is required.
Currently, segmentation is done on image sequences pertaining only to any one single modality such as MRI brain or MRI breast. For this, bio inspired techniques such as Particle Swarm Optimization (PSO), Cuckoo Search Optimization (CSO), and Ant Bee Colony (ABC), are used. The proposed system produces better segmentation results in multimodal (specifically, MRI brain and MRI breast) with various image sequences and yields better segmentation and classification accuracy. This is achieved by integrating Kernel Possibilistic C Means (KPCM), a technique to calculate the kernel-induced distance metrics between data spaces, with PSO that provides the best optimal solution with less number of iterations to increase the efficiency of segmentation and classification of tumors. The approach also involves the removal of irrelevant pixels from the segmented output using morphological reconstruction filter. This enhanced output (feature set) is fed into Support Vector Machine classifier one of the supervised classification techniques for classifying the tumors into normal, benign, and malignant.
The major contributions of this approach in image segmentation are:
- •
Integration of a clustering method (KPCM) with a soft computing technique (PSO);
- •
Deployment of a classifier (SVM) for tumour classification;
- •
Ability to process image sequences of two different modalities (MRI brain and MRI breast) with less computational time.
The rest of the paper organized as follows: section II briefs the related work, Section III deals with image acquiring, types of datasets used, and working principle of KPCM-PSO with Morphological reconstruction filters and support Vector Machine (SVM) architecture in detail. Section III briefly discusses the experimental results and its performance measures and in Section IV conclusion and limitation of the proposed work is stated in a short note.