Integrating Active Learning and Transfer Learning for Carotid Intima-Media Thickness Video Interpretation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Health

Background:

• Cardiovascular disease (CVD) is a leading cause of death, necessitating effective prevention strategies.
• Carotid intima-media thickness (CIMT) imaging is a valuable noninvasive tool for identifying individuals at risk of CVD.
• Automating CIMT video interpretation using deep learning is hindered by the lack of large, annotated datasets and the high cost of manual annotation.

Purpose of the Study:

• To develop a cost-effective method for annotating CIMT videos to facilitate deep learning-based interpretation.
• To introduce a novel framework that integrates active learning and transfer learning for efficient annotation.
• To significantly reduce the time and expertise required for CIMT video annotation.

Main Methods:

• Proposed a new concept of Annotation Unit (AU) simplifying CIMT video annotation to six mouse clicks.
• Developed the Active Fine-Tuning (AFT) algorithm, integrating active and transfer learning with pre-trained Convolutional Neural Networks (CNNs).
• AFT iteratively selects informative unannotated AUs for annotation and fine-tunes the CNN to enhance performance.

Main Results:

• The AFT method demonstrated a substantial reduction in annotation costs, exceeding 81% compared to training from scratch and over 50% compared to random selection.
• The proposed Annotation Unit (AU) simplifies the complex task of CIMT video annotation.
• The AFT algorithm's active, continuous learning capability significantly improves efficiency.

Conclusions:

• The AFT method offers a highly efficient and cost-effective solution for annotating CIMT videos, overcoming a major bottleneck in developing automated CVD risk assessment tools.
• This approach accelerates the development of deep learning models for clinical applications in cardiovascular imaging.
• The study highlights the potential of integrating active learning and transfer learning for efficient medical image annotation.