Thyroid Ultrasound Image Detection and Classification Using Metaheuristic Optimization-based Classifier

Early detection of thyroid nodules enables personalized treatments, reducing mortality and morbidity rates.But, obtaining sufficient ultrasound images is challenging, often resulting in insufficient data which hinders the effective development of innovative techniques.In this research, the Improved Gazelle Optimization Algorithm with Extreme Gradient Boosting (IGOA -XGBoost) is proposed for the classification of thyroid ultrasound images.The CNN is used to extract relevant features as it easily identifies the ultrasound images.The CNN output is given to the XGBoost classifier to classify the input image into benign or malignant classes.Logistic mapping is initialized in traditional GOA to optimize the hyperparameters of XGBoost and achieve better global search and large convergence speed, further enabling the best possible performance.The bias field estimation and correction technique is used as preprocessing to enhance the image quality.The U-Net is employed for segmenting affected regions in the ultrasound image.The IGOA-XGBoost performance is estimated with accuracy, recall, precision, f1-score and specificity for the DDTI dataset.It achieves 99.62% accuracy, 99.35% recall, 99.57% precision, 99.15% f1-score and 98.24% specificity, as opposed compared to the existing techniques like Xception and QuCNet.