Background: Thyroid nodules are commonly encountered in clinical practice, and fine-needle aspiration cytology (FNAC) remains the primary diagnostic tool for differentiating benign from malignant thyroid lesions. However, interpretation of thyroid cytology smears may be challenging due to overlapping morphological features, indeterminate categories, and interobserver variability. Deep learning-based image analysis has emerged as a promising tool for automated classification of cytology smears.

Objective: To evaluate the diagnostic utility of deep learning-assisted classification of thyroid cytology smears and assess its performance in differentiating benign, indeterminate, and malignant thyroid lesions.

Materials and Methods: A retrospective observational study was conducted using 3,500 digitized thyroid cytology smear images obtained from 420 FNAC cases. Images were categorized according to the Bethesda System for Reporting Thyroid Cytopathology. A convolutional neural network model was trained to identify cytomorphological patterns and classify smears into benign, suspicious, and malignant categories. Model performance was compared with cytopathologist diagnosis and histopathological follow-up.

Results: The deep learning model achieved an overall accuracy of 94.2%, sensitivity of 92.8%, specificity of 95.6%, positive predictive value of 91.4%, and negative predictive value of 96.3%. The highest accuracy was observed for benign colloid nodules and papillary thyroid carcinoma. Lower accuracy was noted in follicular-patterned lesions and indeterminate Bethesda categories.

Conclusion: Deep learning-assisted analysis of thyroid cytology smears provides objective, reproducible, and efficient classification support. It can improve diagnostic consistency, reduce workload, and assist in identifying malignant features, particularly when used as an adjunct to expert cytopathological evaluation.