AI model shows promise in thyroid nodule classification but struggles with dataset shift

Researchers have developed a deep ensemble model for classifying thyroid nodules in ultrasound images, aiming to provide calibrated probabilities and uncertainty estimates for clinical decision support. The model, utilizing ConvNeXt-Tiny with attention mechanisms, was trained and validated on the TN5000 dataset, achieving strong internal performance metrics. However, when tested on an independent dataset (TN3K) exhibiting dataset shift, the model's performance and calibration significantly decreased, highlighting challenges in external threshold transportability. AI

IMPACT This research highlights the need for robust model calibration and validation across different datasets for reliable AI deployment in medical imaging.

RANK_REASON The cluster contains an academic paper detailing a new model and its evaluation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Calibrated Selective Prediction Using Deep Ensembles for ROI-Based Thyroid Nodule Ultrasound Classification Under Dataset Shift: A Retrospective Evaluation
• ConvNeXt-Tiny
• Sadibul Hasan Sadib
• TN3K
• TN5000

• paper
• model release
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →