A Novel Framework of Multi-Scale Feature Fusion-Based Refinement Network for Thyroid Nodule Segmentation Using Improved Heuristic Algorithm

The thyroid is a necessary and important endocrine organ in the human body, which maintains the development, growth, and metabolism of the human body by producing thyroid hormones. To perform its functions in the human body effectively, thyroid hormones are essential. The “incidence of thyroid nodules” has been increasing in recent years. Thyroid nodules affect the normal life of individuals and sometimes result in thyroid tumors. To detect and treat thyroid tumors at the initial stages, it is necessary to determine the nodule precisely. Thyroid nodules are abnormal masses created by lesions that can impact the primary symptoms of thyroid abnormalities. For the timely diagnosis of thyroid cancer, thyroid nodule segmentation is a significant step. However, ultrasound image analysis is a challenging and tiresome task, and it is of great importance for thyroid cancer detection. Thyroid nodule automatic segmentation is complex due to the proximity of the background heterogeneity and its appearance; it cannot achieve promising segmentation outcomes because the task struggles to attain accurate boundaries. Hence, this work introduces a novel framework for thyroid nodule segmentation based on the adaptive multi-scale feature fusion technique. Initially, the patient’s medical images are collected from benchmark data resources. The gathered images are then subjected to the Adaptive Multi-Scale Feature Fusion-based Refinement Network (AMFF-RNet) for segmentation. The developed method effectively segments the thyroid nodule abnormalities. To improve the functionality of the model, the model’s parameters are tuned by the Iteration-based Position in Energy Valley Optimizer (IPEVO). The efficacy of the designed thyroid nodule segmentation system is analyzed by the obtained results. The results indicate that the developed method based on the refinement network outperformed traditional models.