Detection of Malnutrition in Children Through Visual Diagnosis of Pediatric Images Using Deep Learning
Early identification of nutritional deficiencies in children is paramount for effective health intervention and long-term recovery. This project presents a non-invasive, AI-driven approach to screening pediatric malnutrition by analyzing full-body imagery through Deep Learning. Developed as a user-friendly web application using Python and the Flask framework, the system provides health workers and clinical staff with an automated tool to classify pediatric cases into "Nourished" or "Malnourished" categories.
The architecture of this diagnostic system relies on two powerful Convolutional Neural Networks (CNNs): ResNet152V2 and MobileNet. Our researchers trained these models on a balanced dataset of over 2,000 pediatric images, utilizing advanced preprocessing techniques like data augmentation and resizing to ensure high generalization across different environments.
According to our performance analysis, the ResNet152V2 model delivered an exceptional 96% test accuracy, while the MobileNet variant provided a highly efficient 95% accuracy with faster processing speeds. By converting visual data into actionable medical insights, this project serves as a scalable solution for nutritional monitoring, particularly in rural or resource-constrained regions where traditional diagnostic equipment may be unavailable.