The pipeline of most segmentation algorithms begins and ends with a curated dataset. In medical imaging, these datasets will be the result of significant effort from clinicians, with each contour representing the result of a sequence of informed decisions. With so much effort and experience involved in the creation of these datasets, we must consider what information about annotation uncertainty is lost over the course of the pipeline. Annotation uncertainty is an inherent property of any dataset containing expert interpretations, and is especially important to quantify in a safety-critical domain such as medical imaging. In this Perspective, we examine the consequences of losing information concerning annotation uncertainty, not only for dataset representation, but also for the reliability of model evaluation. We describe this as annotation-induced evaluation uncertainty: variation in reported model performance arising from the particular annotation used for evaluation, when multiple distinct expert annotations may be plausible. By contrast, considerable research attention has been given to model uncertainty over the last decade, including the development of probabilistic representations for predicted segmentations that allow nuanced spatial information to be included in uncertainty quantification. Comparable representations for annotation uncertainty should complement these developments, enabling evaluation to account for uncertainty not only in the prediction, but also in the annotation. This would allow more reliable and informative evaluations of medical segmentation models.
Artificial intelligence (AI)-enabled imaging systems are often evaluated on individual images, although many applications require accurate measurement of change across visits. Errors that persist across visits can cancel when measurements are subtracted; errors that vary can create apparent change. We combine established measurement principles in a framework that evaluates error in measured change, the proportion of cases yielding both measurements (pair availability), and the acquisition and processing conditions tested. Two synthetic lesion-area experiments used exact reference geometry. The repeated-camera experiment used 72 lesions divided between no change and 10% area growth, with 960 combinations of acquisition and processing settings per arm using at least one trained AI model. The acquisition-resampling experiment varied camera pose, distance, and illumination between visits for the same 36 unchanged lesions. Repeating camera and lighting conditions favored similar errors across visits: assuming independent errors overestimated no-change paired-error standard deviation (SD) by a median factor of 6.4. In the no-change arm, rankings by pooled single-image error SD and paired-change error SD were correlated (Spearman correlation, 0.724), but reversed in 23.2% of unequal comparisons. Reversals occurred in 23.8% of comparisons with a 1.5-2-fold single-image SD advantage and 13.8% with a 2-3-fold advantage; nearly tied rankings were unstable when lesions were resampled. In the repeated-camera experiment, both measurements were obtained for 95.3% of 8,640 scheduled pairs; all final failures arose at marker detection. Compared with matched acquisition, acquisition resampling increased paired-change error SD in all 48 configurations using a trained AI model, by a median factor of 6.2, while single-image error SD changed much less. These findings support evaluating change error and measurement failures directly. The numerical effects depend on the synthetic targets and acquisition conditions and do not establish clinical performance.
Mara Rubia Marques, Anderson Camargo Moreira, Iara Frangiotti Mantovani, Pedro Vale de Azevedo Brito, Isabela Cristina Gomes de Souza Nascimento, Celso Peres Fernandes, Fernanda Cristina Alcantara dos Santos
Age-related Macular Degeneration (AMD) is a leading cause of irreversible vision loss among the elderly population worldwide, making early detection and reliable severity assessment essential for timely intervention and prevention of severe visual impairment. This study presents a novel machine-learning–based framework for automated AMD detection and exploratory severity grading using Optical Coherence Tomography (OCT) images. The proposed approach employs deep learning architectures, including ResNet and VGG, trained on publicly available datasets. An unsupervised learning strategy is introduced for dry AMD exploratory severity grading through feature extraction based on drusen size. Advanced AMD manifestations, such as geographic atrophy and choroidal neovascularization, are identified using clinically relevant retinal features derived from OCT images. The proposed detection framework achieves high accuracy (98%) on the primary dataset and demonstrates robust generalization with an accuracy of 83% for cross-dataset evaluation using transfer learning. Furthermore, the unsupervised severity grading model attains a silhouette score close to 0.5, reflecting meaningful cluster structure consistent with the continuous nature of drusen progression. Overall, the proposed framework offers a scalable and automated solution for AMD detection and exploratory severity grading, with the potential to support clinical decision–making and enhance personalized disease management.
Data clustering is a key tool in medical image analysis and classification. Traditional methods such as k -means often require prior knowledge of the number of clusters and may perform poorly in complex, high-dimensional medical datasets. This study develops and evaluates a novel segmentation method based on Dynamic Quantum Clustering (DQC) for brain tumor MRI analysis. The method is first applied to contrast-enhanced T 1 -weighted MRI and compared with conventional algorithms such as k -means. DQC is then extended to multi-modality MRI datasets using two approaches: independent analysis of each modality and joint analysis of fused multi-modality data. Algorithm parameters are optimized to achieve rapid convergence of the clustering dynamics. Results show that, in single-modality MRI, DQC performs comparably or better than k -means while eliminating the need to predefine the number of clusters. In multi-modality applications, DQC provides effective segmentation for both independent and fused datasets. These findings demonstrate that Dynamic Quantum Clustering is a flexible and efficient approach for medical image segmentation, with strong potential for broader applications in medical image analysis.
