A Hybrid Coot Based CNN Model for Thyroid Cancer Detection

Abstract

Thyroid cancer is one of the most common endocrine malignancies, and early diagnosis is crucial for effective treatment. Fine-needle aspiration biopsy (FNAB) is widely used for diagnosis, but its accuracy depends on expert interpretation, which can be subjective. Recent advances in deep learning, particularly Convolutional Neural Networks (CNNs), have shown promise in automating and improving diagnostic accuracy from biopsy images. However, optimizing CNN architectures remains a challenge, as selecting the best layer parameters significantly impacts performance. Traditional approaches for selecting optimal CNN parameters often depend on exhaustive trial-and-error methods, which are computationally expensive and do not always yield globally optimal solutions. This process is both time-consuming and does not guarantee the precise attainment of an optimal CNN model. In this study, a novel approach is introduced to optimize CNN parameters by utilizing the COOT Metaheuristic Optimization Algorithm, proposing a new model named COOT-CNN for thyroid cancer detection. The COOT algorithm, formulated in 2021 and inspired by the behavioral optimization of waterfowl, is employed in this research to determine the optimal layers and parameters of the CNN model for thyroid cancer diagnosis. This method facilitates efficient optimization of layer parameters through a well-designed coding scheme. The model's efficacy is assessed using thyroid fine needle aspiration biopsy data, categorized into two classes. Performance of the proposed approach is evaluated by comparing it with traditional CNN, Particle Swarm Optimization-based CNN model (PSO-CNN), and Gray Wolf Optimization-based CNN model (GWO-CNN). The proposed model was found to achieve higher accuracy compared to conventional CNN, PSO-CNN, and GWO-CNN models.