Vitamin Deficiency Detection Using CNN-Based Image Analysis of Skin, Tongue and Nails

Vitamin deficiency is a widespread global health issue that affects millions, often leading to severe physiological and dermatological complications. Early detection is essential for timely intervention to prevent the progression of these deficiencies into more serious health concerns. Traditional diagnostic methods, such as invasive blood tests and extensive clinical examinations, are effective but often time-consuming, costly, and uncomfortable for patients. This study explores the potential of non-invasive diagnostic tools leveraging Convolutional Neural Networks (CNNs) for detecting vitamin deficiencies through the analysis of images of the skin, tongue, and nails. The methodology involves the collection of a comprehensive dataset of images exhibiting visible symptoms of vitamin deficiencies from online sources. preprocessing techniques focused on Wiener filtering, which was applied to enhance image clarity and reduce noise. The Wiener filter was selected due to its superior performance ensuring high quality inputs for subsequent analysis. Image segmentation was conducted using Otsu's method to isolate regions of interest, such as discoloration or texture changes indicative of deficiencies. This segmentation step was essential in focusing the analysis on the most relevant parts of the images, improving the reliability of feature extraction. A customized Convolutional Neural Network (CNN) architecture was developed to analyse the pre-processed images. The model was designed with several convolutional layers, pooling layers, and fully connected layers to capture and classify patterns associated with vitamin deficiencies effectively. Hyperparameter tuning was performed to optimize the model's performance, including adjustments to learning rate, batch size, number of epochs, and the architecture's depth.