History and Objectives: The thyroid gland, which is one of the major endocrine organs in the human body, plays a crucial role in regulating daily metabolism. Death rates from thyroid disorders are decreased by early identification. Radiologists and pathologists typically diagnose thyroid illness, and this process strongly relies on their training and knowledge. This study reveals that deep learning-powered algorithms successfully identify thyroid problems automatically, supporting doctors' diagnostic choices and lowering the incidence of human false-positive diagnoses. The present study is a pioneering effort in the field, as it is the first of its kind to employ two preoperative medical imaging modalities for the purpose of multi-classifying thyroid disease categories. The mentioned elements encompass adenoma, a benign condition characterized by the presence of cystic structures and many nodules, as well as thyroiditis, a condition involving inflammation of the thyroid gland. Additionally, the term "normal" is included. This paper presents a diagnostic model for thyroid disease utilizing a cutting-edge deep convolutional neural network (CNN) architecture. The model aims to distinguish between different types of thyroid illnesses. The findings of the study are as follows: The model demonstrates high performance in both categories of medical images, achieving accuracy scores of 0.972 for computed tomography (CT) scans and 0.942 for ultrasound images. The experimental findings underscore the appropriateness of the chosen convolutional neural network (CNN) for both visual modalities, hence emphasizing the potential clinical uses of the deep learning model.