Currently, thyroid cancer and thyroid nodules disorders are increasing globally. The diagnosis of these conditions relies on the development of medical technology. Current methods often suffer from the overfitting issue due to a small dataset and a lack of generalizability to diverse clinical settings. Some of the traditional methods encounter challenges with false positive and false negative rates, which affects the performance of the model. To overcome these challenges, a novel module called Pyramid-SpinalNet (Py-SpinalNet) has been developed for thyroid cancer classification. From the given database, the image is pre-processed through the Wiener filter. After this, 3D-UNet is employed for nodule segmentation. In addition, key features are derived through the process of feature extraction. Eventually, the Py-SpinalNet is used for the classification of thyroid cancer. The Py-SpinalNet is developed by merging PyramidNet and SpinalNet. Here, Accuracy, True Positive Rate (TPR), and True Negative Rate (TNR) are the metrics employed for Py-SpinalNet acquired 91.9, 90.9 and 92.8%. The Py-SpinalNet model can accurately detect thyroid cancer at the early stage, thereby minimizing both false-positive and false-negative rates. Thus, it offers a more efficient and reliable classification of thyroid cancer.