A novel fern-like lines detection using a hybrid of pre-trained convolutional neural network model and Frangi filter

Abstract

Full ferning is the peak of the formation of a salt crystallization line pattern shaped like a fern tree in a woman's saliva at the time of ovulation.The main problem in this study is how to detect the shape of the salivary ferning line patterns that are transparent, irregular and the surface lighting is uneven. This study aims to detect transparent and irregular lines on the salivary ferning surface using a comparison of 15 pre-trained convolutional neural network models. To detect fern-like lines on transparent and irregular layers, a pre-processing stage using the Frangi filter is required. The pre-trained convolutional neural network model is a promising framework with high precision and accuracy for detecting fern-like lines in salivary ferning. The results of this study using the fixed learning rate model ResNet50 showed the best performance with an error rate of 4.37% and an accuracy of 95.63%. Meanwhile, in implementing the automatic learning rate, ResNet18 achieved the best results with an error rate of 1.99% and an accuracy of 98.01%. The results of visual detection of fern-like lines in salivary ferning using a patch size of 34×34 pixels indicate that the ResNet34 model gave the best appearance.

A novel women's ovulation prediction through salivary ferning using the box counting and deep learning

“…In this study, in addition to calculating the fractal dimensions of salivary ferning using pixel counting and box counting, the process of calculating fern-like lines detection in salivary ferning using computer vision or patch windows is also carried out in computing in deep learning. Based on previous research [11] that compared fifteen pre-trained convolutional neural network (CNN) models to detect and visualize the shape of fern-like lines patterns from salivary ferning images with a size of 578x814 pixels, ResNet50 has the best performance with an error rate of 4.37% and an accuracy of 95.63%. The prediction of ovulation time for each female volunteer was carried out in each month of her menstrual cycle through the salivary ferning image dataset using the three methods mentioned above.…”