Covariance weighted distance metrics for optical diagnosis of cancer

Classification of normal and various graded cancer tissues requires a robust distance measure to account for the problems of background noise, source separation and outliers which are inherent to elastic scattering spectroscopy data. It must also have the interpretations for the variation in correlations existing in refractive index fluctuations due to inhomogeneity between healthy and different grades of cancerous tissues. In this contribution, we propose the amalgamation of the L1 distance family and Mahalanobis distance metrics to account for problems mentioned above. The proposed metric for the special case of Manhattan and Chebyshev with Mahalanobis metric has been shown. The efficacy of the proposed distance measure based classification to discriminate the normal and graded cancer tissues with K-NN classifier have been done. Classification accuracy of 93.75%, with sensitivity of 100%, and specificity of 91.94%, validates the suitability of the proposed methodology for pre-cancer detection.