Fuzzy-Label Weighted Deep Learning Classification for CT Image Quality Evaluation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Quality Assessment

Background:

• Assessing computed tomography (CT) image quality is crucial for accurate medical diagnoses.
• Current methods may lack the nuance to handle variations in image quality related to radiation dose.
• Deep learning offers potential but requires robust training data and methods.

Purpose of the Study:

• To develop a deep learning-based image classification approach for assessing CT image quality.
• To introduce a "fuzzy-label" concept to reflect annotation confidence for improved training.
• To determine if CT images pass quality assessment (QA) at specific radiation doses.

Main Methods:

• Proposed a fuzzy-label weighting method to enhance deep learning model training.
• Introduced the concept of "fuzzy-label" to quantify ground-truth annotation confidence.
• Developed an ensemble/assimilation method using CT windowing (window-width/window-length) for image-level quality assessment.

Main Results:

• The fuzzy-label weighted deep learning approach significantly improved CT image classification accuracy.
• The proposed method outperformed traditional baseline image classification approaches.
• The model demonstrated effectiveness even when trained with annotations from a single annotator.

Conclusions:

• Fuzzy-label weighting is an effective strategy for training deep learning models in image quality assessment.
• The proposed method provides a more robust and accurate way to assess CT image quality.
• This approach has the potential to improve diagnostic reliability in medical imaging.