Deep learning-based dual optimization framework for accurate thyroid disease diagnosis using CNN architectures

Abstract

Thyroid diseases, including hypothyroidism, hyperthyroidism, thyroid nodules, thyroiditis, and thyroid cancer, are among the most prevalent endocrine disorders, posing significant health risks, which need to be diagnosed and treated promptly. Traditional diagnostic approaches, reliant on manual interpretation of medical images, are time-consuming and prone to errors. This study introduces a novel deep learning framework utilizing advanced Convolutional Neural Networks (CNNs), specifically modified ResNet and InceptionV3 architectures, to improve the accuracy and efficiency of thyroid disease diagnosis. We present Dual-OptNet, a new hybrid deep learning architecture that effectively merges skip connections of ResNet with multi-scale feature extraction based on InceptionV3 for lung classification tasks. Dual-OptNet shows the most accurate and generalizability results in classifying the thyroid disease with an average and best classification accuracy of 97% from a dual-step optimized using Adam and SGD. Future work will focus on developing a real-time classification tool to demonstrate the potential utility of this model in a clinical context. Future work will also focus on enhancing the dataset to cover a wider range of uncommon thyroid cases, and incorporating explainable AI methods, so that the model decisions are more interpretable. Further research will also explore real-time ultrasound analysis and multi-modal data integration, such as combining medical images with patient history, to enhance diagnostic accuracy. Deploying the system in clinical environments will be key to validating its impact and scalability, ultimately contributing to more efficient and accurate healthcare solutions