Related Experiment Video
Updated: Jun 20, 2025
09:37
Combining Reflectance Confocal Microscopy with Optical Coherence Tomography for Noninvasive Diagnosis of Skin Cancers via Image Acquisition
Published on: August 18, 2022
2.3K
Image Quality Assessment Using Convolutional Neural Network in Clinical Skin Images
Hyeon Ki Jeong1, Christine Park2, Simon W Jiang2
1Department of Biostatistics & Bioinformatics, Duke University School of Medicine, Durham, North Carolina, USA.
Summary
A deep learning model was developed to assess clinical image quality from patients and physicians. This tool improves diagnostic accuracy and clinical workflow efficiency by distinguishing good from bad quality images.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Clinical Diagnostics
Background:
- Suboptimal image quality frequently hinders clinical evaluation.
- Accurate assessment of medical images is crucial for effective patient care.
- Current methods for image quality assessment can be time-consuming and subjective.
Purpose of the Study:
- To develop and validate a deep learning-based tool for automated image quality analysis.
- To assess the quality of images derived from both patients and primary care physicians.
- To improve the efficiency and reliability of image quality assessment in clinical workflows.
Main Methods:
- A VGG16 deep learning model was fine-tuned using a dataset of 4,000+ patient- and primary care physician-derived images.
- Ordinal image quality labels were converted to binary (good vs. bad) using a majority vote.
- Optimal classification thresholds were determined using Youden's index to maximize sensitivity and specificity.
Main Results:
- The model achieved an area under the curve (AUC) of 0.885 on the test set, with high sensitivity (0.829) and specificity (0.784).
- Independent validation on 300 images confirmed robust performance with an AUC of 0.864.
- The tool demonstrated high positive predictive value (0.906 on test set, 0.959 on validation set), indicating reliable identification of good quality images.
Conclusions:
- A practical deep learning approach effectively assesses clinical image quality, enhancing diagnostic workflows.
- The tool offers improved workload and efficiency for clinical teams, despite potential needs for image retakes.
- Automated image quality analysis can significantly benefit healthcare by ensuring reliable diagnostic inputs.
