AI-Assisted Diagnosis of Neonatal Ear Anomalies Using ResNet50 CBAM and CNN Transformer Models

This project presents an AI-assisted framework for image classification of neonatal ear anomalies. The system integrates ResNet50 enhanced with Convolutional Block Attention Module (CBAM) and CNN Transformer architectures to classify auricle images into nine categories, including Normal, Prominent ear, Stahl ear, and other deformities. Using the BabyEar4k dataset, images are preprocessed through resizing, normalization, and quality filtering before input to the models. The framework combines local feature extraction with attention mechanisms to focus on critical regions while capturing global contextual patterns. Models are trained using cross-entropy loss and optimized for accuracy through learning rate scheduling. Performance is evaluated using metrics such as accuracy, precision, recall, F1-score, and confusion matrices to provide a comprehensive assessment of classification performance. The system is implemented as a web application with Flask for backend operations and HTML/CSS/JavaScript for frontend interaction. Users can upload images, receive predictions with associated probabilities, and store results for review. The proposed approach demonstrates high accuracy and robustness in classifying neonatal ear images, providing a standardized tool for anomaly detection.

Keywords: Neonatal ear anomalies, Image classification, ResNet50, CBAM, CNN Transformer, BabyEar4k dataset, Deep learning, Attention mechanism, Auricle deformities, AI-assisted diagnosis