A progressive increase in the demand for more nimble, precise, and accurate medical image assessments has been observed. Throughout the preceding years, technology—especially deep learning (DL)—has developed at an astounding pace, resulting in various applications in helping medical specialists to massively speed up the process, one of which is the diagnosis of thyroid nodules in ultrasound images. In this paper, we developed a hybrid DL model that combines a CNN-based network, Inception-v4, with Grid Search Optimization (GSO) technique to detect and classify thyroid nodules in ultrasound images into two classes, namely, benign and malignant. This approach helps improve the performance of Inception-v4 by choosing the optimal set of hyperparameters with GSO. We utilized the publicly accessible Thyroid Digital Image Database (TDID) to train and assess the performance of the proposed method based on four evaluation metrics: Accuracy, Specificity, Sensitivity, and F-measure, obtaining 98.97%, 95.35%, 99.75%, and 99.37%, respectively. These experimental results validate the model’s efficacy in automating thyroid ultrasound image analysis, thereby streamlining the diagnostic process.