Ensemble-Based Deep Learning Approach for Characterization of Thyroid Nodules in Ultrasound Imaging

This paper addresses the challenge of distinguishing between benign and malignant thyroid nodules using ultrasound (US) imaging, which can be particularly difficult for less experienced clinicians. To aid in this diagnostic process, the paper proposes an ensemble-based classification model that leverages pretrained convolutional neural network (CNN) models. The model combines the outputs of fine-tuned pretrained networks using various ensemble learning strategies, including two-level stacking, three-level stacking, and weighted average ensemble (WAE). The proposed model is evaluated using the DDTI dataset, demonstrating that the ensemble approach offers more promising results and superior prediction accuracy compared to individual models. These findings highlight the potential of these models to assist healthcare professionals in diagnosing thyroid nodules more accurately, thereby enabling timely and appropriate treatment and increasing patient survival rates.