Power spectra based classification of cancerous nevoscope skin images

This paper describes a new method to discriminate between benign and malignant skin cancer samples obtained from the nevoscope which is one of the most commonly used skin imaging apparatus amidst an array of others including the electron microscope and the spectrometer. Although there have been various approaches in the literature proposed for skin cancer detection, they lack from not being very robust to noise and variations in the input images and the effective computational cost. In particular, the work done in [1] makes use of basic feature extraction and an expert system to help in differentiating the skin samples. It involves extraction of many different features and human intervention. In this paper we propose a new approach to skin cancer classification problem based on power spectrum estimation in the frequency domain and demonstrates that significant changes between the two classes can be derived. The periodogram is a means for the spectrum estimation and effectively utilized here in. In the implementation, periodograms of sampled windows of the two classes, benign and malignant skin lesions, are compared to classify them in their respective classes. This is achieved by observing the variations in different window sizes for sampling and determining the most discriminative window size, respectively. The experimental results show that power spectra based classification of cancerous nevoscope skin images is an effective means (i.e almost 97% accurate classification) of non-invasively detecting skin cancer with potential applications in biomedical imaging and related technologies(eg. preventive health care, biopsy, dermatology, etc.).