Hybrid Efficient Net and Enhanced Recurrent Neural Network Model for Automated Thyroid Abnormality Characterization

Abstract

The most prevalent endocrine disorders are thyroid disorders, and accurate diagnosis is essential to managing and treating the condition. The problem with thyroid disorder diagnosis from ultrasound images is that thyroid nodules have variable appearance, and patient characteristics make diagnosis challenging. This paper proposes a hybrid ERNN-EN framework for automatic thyroid disease classification. In order to improve diagnostic performance, this framework uses an enhanced recurrent neural network (ERNN) to model feature dependencies and EfficientNet to extract spatial features from thyroid gland ultrasound images. The proposed ERNN-EN is evaluated on a public Kaggle dataset on thyroid disease, which contains labeled ultrasound images of the thyroid, along with demographic, clinical, and lab information. The dataset contains labeled ultrasound images of the thyroid along with demographic, clinical, and lab information. The dataset is split into training and testing sets to evaluate the model's performance. The experimental results are compared with RU-Net, ResNet, and DenseNet. The evaluation metrics are F1-score, Precision, Recall, and Accuracy. The proposed ERNN-EN achieves the highest Precision (91.8%), F1-score (93.6%), and Accuracy (95.7%), and achieves better Recall, indicating that the proposed hybrid framework can effectively extract discriminative thyroid characteristics and effectively improve the diagnostic performance. The suggested method can be used as a sophisticated clinical decision support system for precise thyroid disease diagnosis and treatment planning.