Prediction of the Thyroid Cancer using the Deep Learning Based Hybrid Spatial Convolution based LSTM Network (SCBLN)

Abstract

The prediction of thyroid cancer has become a significant task in recent years. Binary classification is frequently the objective, despite the existence of extant diagnostic methods; however, the datasets employed are of limited size, and the results are not validated. Model optimization is the primary focus of current methodologies, while the feature engineering component is not as extensively investigated. In order to circumvent these constraints, this study introduces a method that examines feature engineering for deep learning models. Hashimoto's thyroiditis (primary hypothyroidism), autoimmune thyroiditis (compensated hypothyroidism), binding protein (increased binding protein), and non-thyroidal syndrome (NTIS) (concurrent non-thyroidal illness) can be predicted using the SCBLN Hybrid Spatial Convolution Based LSTM Network(SCBLN) approach. The dataset was initially obtained and can be processed using the normalization strategy. After that, the principal component analysis(PCA) method can be employed to extract the cancer-related features. And finally, the thyroid cancer can be predicted by employing the SCBLN classifier. The overall experimentation was carried out under python environment. According to extensive experiments, the SCBLN classifier achieves the highest accuracy and F1 score, with a 0.99 score. Results indicate that the SCBLN model is a superior option for the detection of thyroid cancer in terms of both computational complexity and accuracy. A comparison of the SCBLN approach's performance to existing studies confirms its superiority.