Detection of papillary thyroid carcinoma nuclei in thyroid fine needle aspiration biopsies by deep learning and machine learning methods

Cancer is a very important group of diseases that occur with the uncontrolled growth of cells in the human body and threaten human health worldwide. Although there are many types of cancer, thyroid cancer is one of the most common types of cancer. Cancers that occur in the thyroid gland can cause serious health problems and can be fatal by disrupting the functioning of the thyroid gland. For this reason, early diagnosis and diagnosis is very important for the treatment of cancer. In recent years, studies on this subject have been carried out continuously. In this study, COOT-CNN model, which is a CNN (Convolutional Neural Network) deep learning model optimised with the metaheuristic COOT algorithm, is proposed to classify thyroid cancer. One of the most important challenges of CNN architectures, which are widely used in classification tasks, is to determine the appropriate layer parameters and build the model. In this thesis, an approach to optimise CNN parameters using the COOT algorithm, which is widely used in the field of optimisation, is presented. This method enables the determination of the most appropriate layer parameters with an effective coding scheme. The proposed COOT-CNN model is applied to the classification problem of two-class thyroid fine needle aspiration biopsy data. The performance of the proposed approach is evaluated by comparing it with popular optimisation algorithms such as Partical Swarm Optimisation (PSO) and Grey Wolf Optimisation (GWO) based models. As a result of the experimental studies, it was observed that the COOT-CNN model using optimised layer parameters outperformed the conventional methods known in the literature in the classification of thyroid cancer by achieving higher accuracy compared to PSO and GWO based models. The COOT-CNN hybrid model achieved an accuracy of 92.59%, the GWO-CNN model 88.89% and the PSO-CNN model 87.88% on the test data. This study has made a significant contribution to the use of machine learning-based methods in clinical applications for detecting thyroid cancer and planning personalised treatments.