Image Quality Assessment Using Convolutional Neural Network in Clinical Skin Images

The study developed a convolutional neural network (CNN) model for assessing the quality of clinical skin images, aiming to provide real-time feedback to improve clinical workflows. The model was trained on 900 images and validated with 120 images, achieving a positive predictive value (PPV) of 0.9, meaning it correctly identified 9 out of 10 good quality images. The model's sensitivity, specificity, and negative predictive value (NPV) were 0.71, 0.84, and 0.52, respectively, with an area under the curve (AUC) of 0.86. When region of interest (ROI) analysis was applied to images that failed the whole image test, sensitivity and NPV improved to 0.82 and 0.64, though specificity decreased to 0.78. The study concludes that while the model effectively distinguishes between useful and poor-quality images, further validation is needed to confirm its applicability in real-world settings.