An Early Thyroid Screening Model Based on Transformer and Secondary Transfer Learning for Chest and Thyroid CT Images

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Thyroid cancer necessitates early diagnosis for improved prognosis.
• Enhanced CT scans offer potential for early thyroid cancer screening.
• Existing CT-based models struggle with data limitations and noise.

Purpose of the Study:

• To develop a robust deep learning model for early thyroid cancer screening using enhanced CT scans.
• To address challenges of limited datasets, small sample sizes, and high noise in CT-based models.

Main Methods:

• Collected enhanced CT scan data from 240 patients.
• Developed a DVT model combining transformer deep neural networks (DNN) and transfer learning.
• Integrated time series data to handle small sample sizes and high noise.

Main Results:

• The DVT model achieved 0.96 prediction accuracy.
• Achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.97.
• Demonstrated high sensitivity (0.94) and specificity (1).

Conclusions:

• The DVT model is a highly effective tool for early thyroid cancer screening.
• This deep learning approach shows potential to aid clinicians and reduce patient costs.
• The study presents a novel CT-based screening method leveraging AI.