Effectiveness of Automatically Curated Dataset in Thyroid Nodules Classification Algorithms Using Deep Learning
The study compared deep learning model performance for classifying thyroid nodules using three datasets: manually curated, automatically curated (full), and automatically curated (accurate subset). The full automatically curated dataset significantly outperformed the manually selected one, showing that using all generated data is better than just a high-accuracy subset. Larger batch sizes also improved performance on the noisy automated data.