Hybrid classification with meta‐heuristic‐enabled optimal feature selection for thyroid detection

Abstract

AbstractThyroid is a widespread disease, affecting most victims. The diagnosis of thyroid remains a complex process, as its detection in patients is highly intricate. Hence, the doctors are needed to be aware of the risk factors and symptoms of the disease. This paper aims to propose a novel thyroid diagnosis scheme, involving three major phases: (a) feature extraction, (b) optimal feature selection, and (c) classification. Initially, the thyroid image and the related data serve as input for diagnosing the disease. In the first phase, the features like, gray level co‐occurrence matrix (GLCM), gray level run length matrix (GLRM), local binary pattern (LBP), local vector pattern (LVP), and local tetra patterns (LTrP) are extracted from the input image. Additionally, the features from data are extracted using Principal Component Analysis (PCA) for resolving the issue of “curse of dimensionality.” The optimal features are then selected using a hybrid optimization approach. The optimally selected features of the image and the data are then subjected to the classification process via convolutional neural network (CNN) and neural network (NN), respectively. Both the classified outputs undergo “AND” binary operation to yield the final classified output. To yield effective classification, the NN model is trained by tuning its weights using the proposed algorithm. Further, this paper introduces a new hybrid algorithm, termed firefly updated lion optimization (SLnO) algorithm (FU‐SLnO), for attaining optimal outcomes. Finally, the efficiency of the proposed work is compared over few other conventional approaches and its superiority is proven.