Microscopic Images Analysis for Saliva Ferning Prediction

Abstract

Monitoring the estrogen level in woman’s body can say a lot about their menstrual cycle and ovulation days. The salt crystallization from saliva forms a certain pattern called ferning which can determine the level of estrogen. There are three important stages: no ferning, partial ferning and full ferning. The proposed system intends to automatically determine the fertile period in a cheap, easy, and precise way. The proposed system includes a hardware part, making a miniaturized microscopic device with high performance for image acquisition, and a software able to analyze the images and displaying real-time results based on human saliva samples. For the image processing part, we propose two approaches. The first approach implies the use of the Bag of Words concept, were the words are patches with different ferning styles. For words description we use: CLAHE for image enhancement, Otsu and Frangi filter for ferning emphasis, Local Binary Pattern (LBP) for feature representation of words. For final classification, a support vector machine (SVM) was used, and we classify the frequency of words in one image (the histogram of words). The second image processing approach uses a convolutional neuronal network (CNN). The experimental results show that the proposed system can be used to classify the ferning stage in salivary images, with an accuracy of 0.96 using the SVM classification.