DEEP LEARNING ALGORITHM FOR CERVICAL CANCER DETECTION BASED ON IMAGES OF OPTOMAGNETIC SPECTRA

Abstract

In order to further investigate performance of Optomagnetic Imaging Spectroscopy in cervical cancer detection, deep learning algorithm has been used for classification of optomagnetic spectra of the samples. Optomagnetic spectra reflect cell properties and based on those properties it is possible to differentiate normal cells from cells showing different levels of dysplasia and cancer cells. In one of the previous research, Optomagnetic imaging spectroscopy has demonstrated high percentages of accuracy, sensitivity and specificity in cervical cancer detection, particularly in the case of binary classification. Somewhat lower accuracy percentages were obtained in the case of four class classification. Compared to the results obtained by conventional machine learning classification algorithms, proposed deep learning algorithm achieves similar accuracy results (80%), greater sensitivity (83.3%), and comparable specificity percentages (78%).