Abstract

Vitamin deficiencies are a major world-wide population-health problem as they affect billions of people in developing nations and developed nations. Such deficiencies are often not diagnosed due to a reliance on invasive lab tests, a lack of awareness and a lack of access to healthcare facilities. It is notable also that many vitamin deficiencies present clear clinical evidence on the external human tissues, such as the skin, tongue, lips, the ocular surface, and nails. These phenotypic cues can also be learnt by using traditional imaging devices, thus enabling the process of non-invasive diagnosis in the process of computational analysis. The research assumed here is a smart image-processing and deep-learning framework that is used to identify vitamin deficiencies by analyzing images of human tissues. The proposed architecture is based on a combination of a classic image-processing mode such as Wiener filtering, curvelet transform, morphological manual operations, and Otsu thresholding (RCNN) to obtain the correct result of classification. A full preprocessing pipeline enhances the fidelity of images, and isolates regions of interest, whereas feature extraction using curvelets allows a fine resolution of edges and textures. The identification of the vitamin-deficient tissue or normal tissue is a high classification accuracy that is reliably achieved by the trained classifier. Experimental measurements indicate that the accuracy of the proposed system can attain a total classification of up to 94%, which is better than the traditional wavelet-based methods. For real-time, mobile-compatible implementation. The system is intended to be a non-invasive option to traditional modes of diagnosis, and it is cost-effective. The results support the practicality of using artificial intelligence and image-processing algorithms to detect vitamin deficiency in its initial stage and conduct screening on a large-scale level as a part of population-wide healthcare.