A Novel Thyroid Nodule Classification using ThyroNet X4 Genesis Model

Abstract

The benign or malignant tumors inside the thyroid gland, known as thyroid nodules, can be solid or fluid-filled, making diagnosis difficult. Radiologists frequently struggle to identify cancerous nodules, which might result in needless biopsies or postponed treatments. Using the Thyroid Imaging Reporting and Data System (TI-RADS) in conjunction with a computer-aided diagnostic framework, this study sought to improve diagnostic accuracy. We evaluated a dataset of 99 patients with 33 benign and 66 malignant lesions. Segmentation was done after the images had been preprocessed with picture binarization and a median filter. We used a three-dimensional convolutional neural network (3D CNN) to extract seven ultrasonic characteristics. A ResNet-based deep learning architecture called the ThyroNet X4 Genesis model was suggested and used to complete the classification assignment. By achieving significant gains in F-measure, recall, specificity, sensitivity, and precision, the system demonstrated its potential to assist radiologists in classifying thyroid nodules with more reliability.

Ähnliche Arbeiten