High-Frequency Ultrasound Dataset for Deep Learning-Based Image Quality Assessment

Area of Science:

• Medical Imaging
• Dermatology
• Artificial Intelligence

Background:

• High-frequency ultrasound (HFUS) is vital for dermatology, enabling detailed skin imaging.
• Deep learning algorithms enhance automated analysis of HFUS images for computer-aided diagnosis.
• Image series analysis improves measurement precision but includes non-informative data (artifacts, noise, poor contact).

Purpose of the Study:

• To develop an automated framework for high-frequency ultrasound image quality assessment in skin imaging.
• To improve the accuracy of computer-aided diagnosis by filtering non-informative ultrasound images.
• To establish a reliable method for selecting high-quality images from acquired series for statistical analysis.

Main Methods:

• Collected and annotated a dataset of 17,425 high-frequency facial skin ultrasound images.
• Developed a framework using a deep convolutional neural network (CNN) and a fuzzy reasoning system for image quality assessment.
• Compared different binary and multi-class image analysis approaches based on the VGG-16 model.

Main Results:

• The automated system achieved 91.7% accuracy in binary classification (correct/not correct).
• Multi-class image analysis reached 82.3% accuracy.
• The framework effectively distinguishes between informative and non-informative ultrasound images.

Conclusions:

• Automated image quality assessment is crucial for reliable computer-aided diagnosis in dermatology using HFUS.
• The proposed deep learning and fuzzy logic framework provides an effective solution for filtering non-informative images.
• This work facilitates more precise and accurate dermatological diagnoses through improved image selection.