Enhancing Thyroid Cancer Diagnosis with Advanced Deep Learning Methods
Bellal Linda, Khemis Kamila, Bendjillali Ridha Ilyas, Kherraf Yamina, Borsali Leila, Bendelhoum Mohammed Sofiane · 2024
The accurate diagnosis of thyroid tumors remains challenging due to their clinical and pathological diversity. This study explores the potential of advanced deep learning models to improve thyroid cancer diagnostics by comparing EfficientNet B4 and MobileNetV3 for histopathological image classification. Both models were fine-tuned using transfer learning on a dataset of 7,272 images, covering Medullary, Papillary, and Vesicular carcinoma types. Performance metrics, including precision, recall, F1-score, and accuracy, were used to assess their effectiveness. Findings indicate that both models are highly effective, with EfficientNet B4 exhibiting superior accuracy, particularly in complex diagnostic scenarios. The study also addresses limitations, such as the dataset’s limited size and variability, which may affect model generalization. Ultimately, these advancements in deep learning could significantly enhance thyroid cancer diagnostics, minimize inter-observer variability, and lead to better patient outcomes.