The Classification and Detection of Cyanosis Images on Lightly and Darkly Pigmented Individual Human Skins Applying Simple CNN and Fine-Tuned VGG16 Models in TensorFlow's Keras API

The target of this research is to evaluate the images classification and detection of cyanosis on lightly and darkly individual human skins using two popular models, simple convolutional neural network (CNN) and Fine-tuned VGG16 in deep neural network TensorFlow's Keras API; and compare the performance of both models based on images that were not incorporated in ImageNet library dataset and data augmentation. Cyanosis is the abnormal bluish stain of skin due to poor circulation or lack of oxygen (the word cyanosis originated from the colour name cyan, which is etymologically derived from the Greek term “kyanos“ meaning “blue”). Peripheral and central cyanosis is indicated by the purple or bluish tint of the feet, face, and hands. The fine-tuned VGG16 model has demonstrated emphatic categorical validation accuracy (82%) than the simple CNN model (79%). On the other hand, the simple CNN model took less time (7.733 minutes) to train on the images that were not incorporated in the ImageNet library than fine-tuned VGG16 model (24.63 minutes). With the limitations of in-vivo and in-silico experiments on the classification and detection of the darkly pigmented individual human skins, it will be veritably beneficial for the dermatologist to apply a deep neural network to prognosis and classify and detect cyanosis in individual human skins at its early stage.