Automated Image Quality Classification In 99mtc-MDP Bone Scintigraphy: An Unsupervised Approach Using PCA And K- Means Clustering

Abstract

Poor-quality images hinder clinical interpretation, while high-quality images are essential for both training and real-world deployment of AI/ML-based diagnostic tools. Reliable image quality assessment (IQA) is therefore crucial for the development of image enhancement algorithms, particularly in nuclear medicine, where diagnostic accuracy relies heavily on image clarity. Traditional subjective IQA methods are time-consuming, costly, and prone to inter-observer variability. Although various supervised learning-based approaches can be effective, they require large annotated datasets, which are often difficult to obtain in clinical practice. In this study, we propose a novel unsupervised IQA framework for 99mTc- MDP bone scintigraphy images using Principal Component Analysis (PCA) and K-means clustering. A dataset of 406 bone scans was processed using PCA for dimensionality reduction, retaining the top 32 principal components. K-means clustering (with k = 3) was then applied to classify images into three quality categories: ‘Poor,’ ‘Good,’ and ‘Excellent.’ The method was validated using 100 randomly selected images labeled by a nuclear medicine physician. The framework achieved an accuracy, sensitivity and specificity of 93%, 93.12% and 96.48% respectively, demonstrating its effectiveness. This unsupervised, interpretable, and data-efficient approach provides a practical solution for automated quality control in nuclear imaging workflows, with the potential to enhance both clinical decision-making and AI-based diagnostics.