An Intelligent Image-Based System for Early Detection of Nutritional Deficiencies Using ResNet and Attention Mechanism

Abstract

 Vitamin and mineral deficiencies are 
widespread health concerns that often remain 
undetected due to reliance on costly and time
consuming laboratory tests. This paper proposes an 
AI-driven, non-invasive diagnostic system that 
utilizes deep learning to identify nutritional 
deficiencies from images of facial skin, eyes, tongue, 
nails, and hair. The model employs transfer learning 
using ResNet152V2 combined with an attention 
mechanism to enhance feature extraction by focusing 
on critical regions associated with deficiency 
symptoms. Data augmentation and regularization 
techniques are applied to improve generalization and 
handle limited datasets. 
The proposed system is capable of detecting multiple 
deficiencies, including Vitamins A, B-complex, C, D, 
E, K, as well as iron and zinc deficiencies, with high 
accuracy, precision, recall, and F1-score. 
Experimental results demonstrate that the model 
effectively captures subtle visual patterns and 
outperforms traditional diagnostic approaches in 
terms of speed and accessibility. The system is 
deployed through a user-friendly interface, enabling 
real-time predictions and supporting early diagnosis. 
This approach provides a cost-effective and scalable 
healthcare solution, particularly beneficial for rural 
and underserved populations, and contributes to 
improved preventive healthcare and timely 
intervention.