Fingernail Image-based Health Assessment Using a Hybrid VGG16 and Random Forest Model

Undetected changes in fingernails consisting of alterations in color, shape, size, or texture can be a reason for unfavorable results for the health of the human. It has been a challenge to detect diseases in the human body at an early stage through fingernail images. Various diseases including kidney disorders, melanoma, anaemia, liver disease, heart disease, and many other diseases, can be diagnosed through the images of fingernails. Early diagnosis of mild cases of a disease can help patients guide additional medical care to stop the progression of the disease. Recently, there has been a substantial amount of interest in applying deep learning to diagnose diseases through fingernail images. There are currently a few algorithms that include Densenet201, InceptionV3, Visual Transform, CNN, KNN, etc. that are used to diagnose diseases through human fingernail images. The limitations of these approaches were limited accuracy in diagnosing the disease through images of the fingernails. In this paper, a hybrid model is built using Random Forest classifier that uses the features extracted from VGG16. The dataset is built manually using 1343 images for the early diagnosis of diseases like kidney disorder, melanoma, and anemia. Moreover, the results obtained from this proposed methodology are compared with previous work done in disease detection through the images of the fingernails, which concluded that the proposed model has better performance than the previous works. The proposed method gives the results the following outcomes in the form of accuracy, precision, recall, and F1 score as 97.02%, 97.02%, 96.84%, and 97.01% respectively.