A RESNET-BASED FRAMEWORK FOR VITAMIN DEFICIENCY DETECTION FROM VISIBLE SYMPTOMS

Geetha Sri Yendrapati , Yasasvi Sai Sree Toleti , Nehapriya Yarlagadda , Mani Kumar Chari Sriramadasu , Sanjay Saikam

Vitamin Deficiency Detection, Deep Learning, ResNet152V2, Transfer Learning, Image Classification, Visible Symptom Analysis, Remote Healthcare.

Vitamin deficiencies represent one of the most common health problems in many countries of the world, especially where access to medical test facilities is poor. These deficiencies could take the form of impaired immunity, a sight problem, skin diseases, anemia, or poor physical development. Conventional techniques for its diagnosis primarily rely on blood tests, which are very accurate but sometimes invasive, expensive, and ill-suited for mass screening or screening in remote areas. This demands the development of other methods of diagnosis that are non-invasive in nature and yet would facilitate early detection. This project is aimed at the development of an AI-driven system for the detection of vitamin deficiencies using images with visible symptoms from eyes, nails, skin, tongue, and hair by deep learning. In this approach, transfer learning applies the ResNet152V2 model for feature extraction, which uses the attention mechanism to help the model focus on most of the significant parts of the image that are associated with deficiency-related visual patterns, improving the detection of very small manifestations of symptoms. A deep learning approach using a custom dataset of images of different body areas and corresponding labels is used to train and test the proposed system. The proposed model will be able to detect multiple vitamin deficiencies and give accurate predictions. The performance of the proposed model tested using common evaluation metrics such as accuracy(91.4%), precision(91%), recall(90.7%), and F1-score(90.8%), which will show the efficiency of the proposed classification model. The proposed model is expected to be used as a supporting diagnostic tool and will work to decrease the dependence on laboratory work. The proposed model will also be non-invasive and will be easily deployable on a large scale.

Volume 14 Issue 3, March-2026

Pages : f826-f834