An Adaptive Low-Rank Modeling-Based Active Learning Method for Medical Image Annotation

Area of Science:

• Medical image analysis
• Machine learning
• Computer vision

Background:

• Active learning efficiently trains models with limited annotated data, crucial for medical imaging where unlabeled data is abundant but annotation is costly.
• Challenges in medical active learning include image noise, large datasets, and diverse imaging modalities, complicating informative example selection.

Purpose of the Study:

• To develop a novel low-rank modeling-based multi-label active learning (LRMMAL) method for optimal medical image classification.
• To address challenges in medical image active learning, including noise, data volume, and modality variety.

Main Methods:

• Developed a low-rank modeling-based multi-label active learning (LRMMAL) method.
• Quantified image noise independently and integrated it into a pool-based sampling process for informative example selection.
• Proposed an automatic adaptive cross-entropy-based parameter determination scheme to optimize sampling.

Main Results:

• The LRMMAL method demonstrated superior performance in selecting informative examples for training classifiers.
• Experimental results on varied medical image datasets confirmed the effectiveness of the proposed method.
• Comparisons with state-of-the-art multi-label active learning methods showed significant improvements.

Conclusions:

• The proposed LRMMAL method effectively addresses key challenges in medical image active learning.
• This approach enhances classifier performance by optimizing the selection of informative training examples.
• The method offers a promising solution for efficient and accurate medical image analysis.