ThyroNet-X4 genesis: an advanced deep learning model for auxiliary diagnosis of thyroid nodules' malignancy

Thyroid nodules are a common endocrine condition, and accurate differentiation between benign and malignant nodules is essential for making appropriate treatment decisions. Traditional ultrasound-based diagnoses often depend on the expertise of physicians, which introduces a risk of misdiagnosis. To address this challenge, this study proposes a novel deep learning model, ThyroNet-X4 Genesis, designed to automatically classify thyroid nodules as benign or malignant. Built on the ResNet architecture, the model enhances feature extraction by incorporating grouped convolutions and using larger convolution kernels, improving its ability to analyze thyroid ultrasound images. The model was trained and validated using publicly available thyroid ultrasound imaging datasets, and its generalization was further tested using an external validation dataset from HanZhong Central Hospital. The ThyroNet-X4 Genesis model achieved 85.55% and 71.70% accuracy on the internal training and validation sets, respectively, and 67.02% accuracy on the external validation set. These results surpass those of other mainstream models, highlighting its potential for clinical use in thyroid nodule classification. This work underscores the growing role of deep learning in thyroid nodule diagnosis and provides a foundation for future research in high-performance medical diagnostic models.