The aim of this validation study was to comprehensively evaluate the performance and generalization capability of a deep learning-based periapical lesion detection algorithm on a clinically representative cone-beam computed tomography (CBCT) dataset and test for non-inferiority. The evaluation involved 195 CBCT images of adult upper and lower jaws, where sensitivity and specificity metrics were calculated for all teeth, stratified by jaw, and stratified by tooth type. Furthermore, each lesion was assigned a periapical index score based on its size to enable a score-based evaluation. Non-inferiority tests were conducted with proportions of 90% for sensitivity and 82% for specificity. The algorithm achieved an overall sensitivity of 86.7% and a specificity of 84.3%. The non-inferiority test indicated the rejection of the null hypothesis for specificity but not for sensitivity. However, when excluding lesions with a periapical index score of one (i.e., very small lesions), the sensitivity improved to 90.4%. Despite the challenges posed by the dataset, the algorithm demonstrated promising results. Nevertheless, further improvements are needed to enhance the algorithm’s robustness, particularly in detecting very small lesions and the handling of artifacts and outliers commonly encountered in real-world clinical scenarios.

Anatomical landmark localization (ALL) plays a crucial role in medical imaging for applications such as therapy planning and surgical interventions. State-ofthe- art deep learning methods for ALL are often trained on small datasets due to the scarcity of large, annotated medical data. This constraint often leads to overfitting on the training dataset, which in turn reduces the model's ability to generalize to unseen data. To address these challenges, we propose a multi-channel generative approach utilizing Flow Matching to synthesize diverse annotated images for data augmentation in ALL tasks. Each synthetically generated sample consists of a medical image paired with a multi-channel heatmap that encodes its landmark configuration, from which the corresponding landmark annotations can be derived. We assess the quality of synthetic image-heatmap pairs automatically using a Statistical Shape Model to evaluate landmark plausibility and compute the Fréchet Inception Distance score to quantify image quality. Our results show that pairs synthesized via Flow Matching exhibit superior quality and diversity compared with those generated by other state-of-the-art generative models like Generative Adversarial Networks or diffusion models. Furthermore, we investigate the effect of integrating synthetic data into the training process of an ALL network. In our experiments, the ALL network trained with Flow Matching-generated data demonstrates improved robustness, particularly in scenarios with limited training data or occlusions, compared with baselines that utilize solely real images or synthetic data from alternative generative models.

The task of localizing distinct anatomical structures in medical image data is an essential prerequisite for several medical applications, such as treatment planning in orthodontics, bone-age estimation, or initialization of segmentation methods in automated image analysis tools. Currently, Anatomical Landmark Localization (ALL) is mainly solved by deep-learning methods, which cannot guarantee robust ALL predictions; there may always be outlier predictions that are far from their ground truth locations due to out-of-distribution inputs. However, these localization outliers are detrimental to the performance of subsequent medical applications that rely on ALL results. The current ALL literature relies heavily on implicit anatomical constraints built into the loss function and network architecture to reduce the risk of anatomically infeasible predictions. However, we argue that in medical imaging, where images are generally acquired in a controlled environment, we should use stronger explicit anatomical constraints to reduce the number of outliers as much as possible. Therefore, we propose the end-to-end trainable Global Anatomical Feasibility Filter and Analysis (GAFFA) method, which uses prior anatomical knowledge estimated from data to explicitly enforce anatomical constraints. GAFFA refines the initial localization results of a U-Net by approximately solving a Markov Random Field (MRF) with a single iteration of the sum-product algorithm in a differentiable manner. Our experiments demonstrate that GAFFA outperforms all other landmark refinement methods investigated in our framework. Moreover, we show that GAFFA is more robust to large outliers than state-of-the-art methods on the studied X-ray hand dataset. We further motivate this claim by visualizing the anatomical constraints used in GAFFA as spatial energy heatmaps, which allowed us to find an annotation error in the hand dataset not previously discussed in the literature.