An effective segmentation algorithm for the hyperspectral cancer images

Cancer is the major reason for mortality worldwide. The chances of recovery can be well improved if it is possible to diagnose the cancer at its early stages. Cancer detection is conventionally done by invasive procedures like biopsy, but it causes lot of discomforts for the patient. Here Hyperspectral Images (HSI) based noninvasive alternative for biopsy is introduced, it is also applicable as a tumor margin assessment during image-guided surgery. HSI is emerged as a promising tool for cancer detection and in tumor margin assessment. The key technique employed here is Marko Random Field (MRF), for depth wise segmentation and it is followed by malignant tissue extraction using Chan-Vese Active Contour method. Here MRF and active contour methods are respectively utilizes local and global features of the subject and the effectiveness of the segmentation at discrete input conditions can be improved by incorporating local optimizations scheme with the segmentation. It helps to make the segmentation more effective and accurate, by compensating the geometrical (ie, containment and exclusion) and distance constraints in the input samples. It is essential because of the spatially-recurring, multi-regional boundary structure of tumor margins. The heterogeneity in the tissue samples can be well defined by the optimization values and it is also be used for tissue classification. Finally feature of the malignant tissue is also extracted for having a clear inference of the subject. The proposed method is compared with Improved watershed transform, Minimum Spanning Forest (MSF) based algorithm and the proposed method is found to be effective than other methods with a high accuracy.