Deep Learning-Based Multilayer Perceptron Model for Classifying Benign and Malignant Thyroid Nodules: Enhancing Diagnostic Accuracy in Clinical Practice

Abstract

This study concludes that the most utilized and diagnostic tool for thyroid nodule assessment is Ultrasound (US), by which the size of the nodule, structure, and other features can be measured. Involving some deep learning systems and particularly Multilayer Perception (MLP), which studies and classifies thyroid nodules as benign and malignant, is illustrated. An MLP-based technique has been performed on a cohort of 847 patients, 1,978 of which are benign and 3,608 are malignant, separated into two batches. The MLP model recorded a remarkable accuracy of 98.91%, indicating precise consistency of the training and validation datasets, with minor fluctuations, hence, solidgeneralization. The model manifested exceptional diagnostic consistency and dependability on account of its admirable learning, steady enhancement of its accuracy, and convergence of the loss curves. With respect to the last-defended deep learning techniques, this MLP model stands out, especially in thyroid nodule classification, assuring a decline in diagnostic-related human mistakes and a faster diagnosis. Future research would enhance the model by involving larger and more diverse datasets and exploring more advanced architectures of neural networks, Convolutional Neural Networks (CNNs), or hybrid models. This MLP model will be integrated into clinical practice to streamline decision-making, enhance patient outcomes, and take one step forward in the evolution of medical image analysis.

Ähnliche Arbeiten