2017 IEEE International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET 2017)

302 / 2017-02-06 17:16:46

A Noninvasive Cancer detection using Hyperspectral Images

12768,12767,12766,12765,6096

Cancer remained as one of the major cause of mortality worldwide. Many studies have been performed on the early detection of cancer using noninvasive or minimally invasive techniques in lieu of traditional excisional biopsy. Immense change in the cancer treatment can be made by early detection, it makes simpler and more effective when diagnosed at an early stage. Here a self sufficient classification method has been introduced that combines both spectral and spatial information on hyperspectral images of an animal tissue in distinguishing cancerous via healthy. An automated algorithm based on an optimal band selection and Minimum Spanning Forest (MSF) algorithm has been proposed for classification of hyperspectral images. The segmentation has been achieved by collecting the features from the selected markers in the optimal band. In optimal band selection, the unnecessary band may get eliminated and selects a range where there is a probability of having abnormality and the markers are being selected using the intensity feature available in the selected band. From these numerous markers available, the problematic region if available may be differentiated using the MSF algorithm. The MSF is utilizing the pixel distribution of every marker. The SVM classifier is finally used to distinguish between the normal or abnormal based on the MSF features. The corresponding MSF based advanced cancer detection scheme is accurate and efficient, making it as a good noninvasive tool. It not only applicable in early cancer detection but also as an interaoperative tool in tumor margins resection.