Purpose: Deep learning methods have shown promising results in the segmentation, and detection of diseases in medical images. However, most methods are trained and tested on data from a single source, modality, organ, or disease type, overlooking the combined potential of other available annotated data. Numerous small annotated medical image datasets from various modalities, organs, and diseases are publicly available. In this work, we aim to leverage the synergistic potential of these datasets to improve performance on unseen data. Approach: To this end, we propose a novel algorithm called MMIS-Net (MultiModal Medical Image Segmentation Network), which features Similarity Fusion blocks that utilize supervision and pixel-wise similarity knowledge selection for feature map fusion. Additionally, to address inconsistent class definitions and label contradictions, we created a one-hot label space to handle classes absent in one dataset but annotated in another. MMIS-Net was trained on 10 datasets encompassing 19 organs across 2 modalities to build a single model. Results: The algorithm was evaluated on the RETOUCH grand challenge hidden test set, outperforming large foundation models for medical image segmentation and other state-of-the-art algorithms. We achieved the best mean Dice score of 0.83 and an absolute volume difference of 0.035 for the fluids segmentation task, as well as a perfect Area Under the Curve of 1 for the fluid detection task. Conclusion: The quantitative results highlight the effectiveness of our proposed model due to the incorporation of Similarity Fusion blocks into the network's backbone for supervision and similarity knowledge selection, and the use of a one-hot label space to address label class inconsistencies and contradictions.

Foundational models are trained on extensive datasets to capture the general trends of a domain. However, in medical imaging, the scarcity of data makes pre-training for every domain, modality, or task challenging. Continual learning offers a solution by fine-tuning a model sequentially on different domains or tasks, enabling it to integrate new knowledge without requiring large datasets for each training phase. In this paper, we propose UNIfied CONtinual Learning for Medical Foundational Models (UNICON), a framework that enables the seamless adaptation of foundation models to diverse domains, tasks, and modalities. Unlike conventional adaptation methods that treat these changes in isolation, UNICON provides a unified, perpetually expandable framework. Through careful integration, we show that foundation models can dynamically expand across imaging modalities, anatomical regions, and clinical objectives without catastrophic forgetting or task interference. Empirically, we validate our approach by adapting a chest CT foundation model initially trained for classification to a prognosis and segmentation task. Our results show improved performance across both additional tasks. Furthermore, we continually incorporated PET scans and achieved a 5\% improvement in Dice score compared to respective baselines. These findings establish that foundation models are not inherently constrained to their initial training scope but can evolve, paving the way toward generalist AI models for medical imaging.

Convolutional neural networks (CNNs) have emerged to reduce clinical resources and standardize auto-contouring of organs-at-risk (OARs). Although CNNs perform adequately for most patients, understanding when the CNN might fail is critical for effective and safe clinical deployment. However, the limitations of CNNs are poorly understood because of their black-box nature. Explainable artificial intelligence (XAI) can expose CNNs' inner mechanisms for classification. Here, we investigate the inner mechanisms of CNNs for segmentation and explore a novel, computational approach to a-priori flag potentially insufficient parotid gland (PG) contours. First, 3D UNets were trained in three PG segmentation situations using (1) synthetic cases; (2) 1925 clinical computed tomography (CT) scans with typical and (3) more consistent contours curated through a previously validated auto-curation step. Then, we generated attribution maps for seven XAI methods, and qualitatively assessed them for congruency between simulated and clinical contours, and how much XAI agreed with expert reasoning. To objectify observations, we explored persistent homology intensity filtrations to capture essential topological characteristics of XAI attributions. Principal component (PC) eigenvalues of Euler characteristic profiles were correlated with spatial agreement (Dice-Sørensen similarity coefficient; DSC). Evaluation was done using sensitivity, specificity and the area under receiver operating characteristic (AUROC) curve on an external AAPM dataset, where as proof-of-principle, we regard the lowest 15% DSC as insufficient. PatternNet attributions (PNet-A) focused on soft-tissue structures, whereas guided backpropagation (GBP) highlighted both soft-tissue and high-density structures (e.g. mandible bone), which was congruent with synthetic situations. Both methods typically had higher/denser activations in better auto-contoured medial and anterior lobes. Curated models produced "cleaner" gradient class-activation mapping (GCAM) attributions. Quantitative analysis showed that PCλ₁ of guided GCAM's (GGCAM) Euler characteristic (EC) profile had good predictive value (sensitivity>0.85, specificity>0.90) of DSC for AAPM cases, with AUROC = 0.66, 0.74, 0.94, 0.83 for GBP, GCAM, GGCAM and PNet-A. For for λ₁ < -1.8e3 of GGCAM's EC-profile, 87% of cases were insufficient. GBP and PNet-A qualitatively agreed most with expert reasoning on directly (structure borders) and indirectly (proxies used for identifying structure borders) important features for PG segmentation. Additionally, this work investigated as proof-of-principle how topological data analysis could be used for quantitative XAI signal analysis to a-priori mark potentially inadequate CNN-segmentations, using only features from inside the predicted PG. This work used PG as a well-understood segmentation paradigm and may extend to target volumes and other organs-at-risk.

The shape and size of the placenta are closely related to fetal development in the second and third trimesters of pregnancy. Accurately segmenting the placental contour in ultrasound images is a challenge because it is limited by image noise, fuzzy boundaries, and tight clinical resources. To address these issues, we propose MCBL-UNet, a novel lightweight segmentation framework that combines the long-range modeling capabilities of Mamba and the local feature extraction advantages of convolutional neural networks (CNNs) to achieve efficient segmentation through multi-information fusion. Based on a compact 6-layer U-Net architecture, MCBL-UNet introduces several key modules: a boundary enhancement module (BEM) to extract fine-grained edge and texture features; a multi-dimensional global context module (MGCM) to capture global semantics and edge information in the deep stages of the encoder and decoder; and a parallel channel spatial attention module (PCSAM) to suppress redundant information in skip connections while enhancing spatial and channel correlations. To further improve feature reconstruction and edge preservation capabilities, we introduce an attention downsampling module (ADM) and a content-aware upsampling module (CUM). MCBL-UNet has achieved excellent segmentation performance on multiple medical ultrasound datasets (placenta, gestational sac, thyroid nodules). Using only 1.31M parameters and 1.26G FLOPs, the model outperforms 13 existing mainstream methods in key indicators such as Dice coefficient and mIoU, showing a perfect balance between high accuracy and low computational cost. This model is not only suitable for resource-constrained clinical environments, but also provides a new idea for introducing the Mamba structure into medical image segmentation.

To develop PerAIDE, an AI-driven system for automated analysis of pulmonary perfusion blood volume (PBV) using dual-energy computed tomography pulmonary angiography (DE-CTPA) in patients with chronic pulmonary thromboembolism (CPE). In this prospective observational study, 32 patients with chronic thromboembolic pulmonary disease (CTEPD) and 151 patients with chronic thromboembolic pulmonary hypertension (CTEPH) were enrolled between January 2022 and July 2024. PerAIDE was developed to automatically quantify three distinct perfusion patterns-normal, reduced, and defective-on DE-CTPA images. Two radiologists independently assessed PBV scores. Follow-up imaging was conducted 3 months after balloon pulmonary angioplasty (BPA). PerAIDE demonstrated high agreement with the radiologists (intraclass correlation coefficient = 0.778) and reduced analysis time significantly (31 ± 3 s vs. 15 ± 4 min, p < 0.001). CTEPH patients had greater perfusion defects than CTEPD (0.35 vs. 0.29, p < 0.001), while reduced perfusion was more prevalent in CTEPD (0.36 vs. 0.30, p < 0.001). Perfusion defects correlated positively with pulmonary vascular resistance (ρ = 0.534) and mean pulmonary artery pressure (ρ = 0.482), and negatively with oxygenation index (ρ = -0.441). PerAIDE effectively differentiated CTEPH from CTEPD (AUC = 0.809, 95% CI: 0.745-0.863). At the 3-month post-BPA, a significant reduction in perfusion defects was observed (0.36 vs. 0.33, p < 0.01). CTEPD and CTEPH exhibit distinct perfusion phenotypes on DE-CTPA. PerAIDE reliably quantifies perfusion abnormalities and correlates strongly with clinical and hemodynamic markers of CPE severity. ClinicalTrials.gov, NCT06526468. Registered 28 August 2024- Retrospectively registered, https://clinicaltrials.gov/study/NCT06526468?cond=NCT06526468&rank=1 . PerAIDE is a dual-energy computed tomography pulmonary angiography (DE-CTPA) AI-driven system that rapidly and accurately assesses perfusion blood volume in patients with chronic pulmonary thromboembolism, effectively distinguishing between CTEPD and CTEPH phenotypes and correlating with disease severity and therapeutic response. Right heart catheterization for definitive diagnosis of chronic pulmonary thromboembolism (CPE) is invasive. PerAIDE-based perfusion defects correlated with disease severity to aid CPE-treatment assessment. CTEPH demonstrates severe perfusion defects, while CTEPD displays predominantly reduced perfusion. PerAIDE employs a U-Net-based adaptive threshold method, which achieves alignment with and faster processing relative to manual evaluation.

Accurate segmentation of lung cancer lesions in computed tomography (CT) is essential for precise diagnosis, personalized therapy planning, and treatment response assessment. While automatic segmentation of the primary lung lesion has been widely studied, the ability to segment multiple lesions per patient remains underexplored. In this study, we address this gap by introducing a novel, automated approach for multi-instance segmentation of lung cancer lesions, leveraging a heterogeneous cohort with real-world multicenter data. We analyzed 1,081 retrospectively collected CT scans with 5,322 annotated lesions (4.92 ± 13.05 lesions per scan). The cohort was stratified into training (n = 868) and testing (n = 213) subsets. We developed an automated three-step pipeline, including thoracic bounding box extraction, multi-instance lesion segmentation, and false positive reduction via a novel multiscale cascade classifier to filter spurious and non-lesion candidates. On the independent test set, our method achieved a Dice similarity coefficient of 76% for segmentation and a lesion detection sensitivity of 85%. When evaluated on an external dataset of 188 real-world cases, it achieved a Dice similarity coefficient of 73%, and a lesion detection sensitivity of 85%. Our approach accurately detected and segmented multiple lung cancer lesions per patient on CT scans, demonstrating robustness across an independent test set and an external real-world dataset. AI-driven segmentation comprehensively captures lesion burden, enhancing lung cancer assessment and disease monitoring KEY POINTS: Automatic multi-instance lung cancer lesion segmentation is underexplored yet crucial for disease assessment. Developed a deep learning-based segmentation pipeline trained on multi-center real-world data, which reached 85% sensitivity at external validation. Thoracic bounding box and false positive reduction techniques improved the pipeline's segmentation performance.

Artificial intelligence (AI) achieves high-quality annotations of radiological images, yet often lacks the robustness required in clinical practice. Interactive annotation starts with an AI-generated delineation, allowing radiologists to refine it with feedback, potentially improving precision and reliability. These techniques have been explored in two-dimensional desktop environments, but are not validated by radiologists or integrated with immersive visualization technologies. We used a Virtual Reality (VR) system to determine whether (1) the annotation quality improves when radiologists can edit the AI annotation and (2) whether the extra work done by editing is worthwhile. We evaluated the clinical feasibility of an interactive VR approach to annotate mandibular and mental foramina on segmented 3D mandibular models. Three experienced dentomaxillofacial radiologists reviewed AI-generated annotations and, when needed, refined them at the voxel level in 3D space through click-based interactions until clinical standards were met. Our results indicate that integrating expert feedback within an immersive VR environment enhances annotation accuracy, improves clinical usability, and offers valuable insights for developing medical image analysis systems incorporating radiologist input. This study is the first to compare the quality of original and interactive AI annotation and to use radiologists' opinions as the measure. More research is needed for generalization.

While lung cancer screening (LCS) reduces lung cancer-related mortality in high-risk individuals, cardiovascular disease (CVD) remains a leading cause of death due to shared risk factors such as smoking and age. Coronary artery calcium (CAC) assessment offers an opportunity for concurrent cardiovascular screening, with higher CAC scores indicating increased CVD risk and mortality. Despite guidelines recommending CAC-scoring on all non-contrast chest CT scans, a lack of standardization leads to underreporting and missed opportunities for preventive care. Routine CAC-scoring in LCS can enable personalized CVD management and reduce unnecessary treatments. However, challenges persist in achieving adequate diagnostic quality with one combined image acquisition for both lung and cardiovascular assessment. Advancements in CT technology have improved CAC quantification on low-dose CT scans. Electron-beam tomography, valued for superior temporal resolution, was replaced by multi-detector CT for better spatial resolution and general usability. Dual-source CT further improved temporal resolution and reduced motion artifacts, making non-gated CT protocols for CAC-assessment possible. Additionally, artificial intelligence-based CAC quantification can reduce the added workload of cardiovascular screening within LCS programs. This review explores recent advancements in cardiac CT technologies that address prior challenges in opportunistic CVD screening and considers key factors for integrating CVD screening into LCS programs, aiming for high-quality standardization in CAC reporting.

Cone-beam computed tomography (CBCT) has become an invaluable imaging modality in dentistry, enabling 3D visualization of teeth and surrounding structures for diagnosis and treatment planning. Automated segmentation of dental structures in CBCT can efficiently assist in identifying pathology (e.g., pulpal or periapical lesions) and facilitate radiation therapy planning in head and neck cancer patients. We describe the DLaBella29 team's approach for the MICCAI 2025 ToothFairy3 Challenge, which involves a deep learning pipeline for multi-class tooth segmentation. We utilized the MONAI Auto3DSeg framework with a 3D SegResNet architecture, trained on a subset of the ToothFairy3 dataset (63 CBCT scans) with 5-fold cross-validation. Key preprocessing steps included image resampling to 0.6 mm isotropic resolution and intensity clipping. We applied an ensemble fusion using Multi-Label STAPLE on the 5-fold predictions to infer a Phase 1 segmentation and then conducted tight cropping around the easily segmented Phase 1 mandible to perform Phase 2 segmentation on the smaller nerve structures. Our method achieved an average Dice of 0.87 on the ToothFairy3 challenge out-of-sample validation set. This paper details the clinical context, data preparation, model development, results of our approach, and discusses the relevance of automated dental segmentation for improving patient care in radiation oncology.

The lumen and external elastic lamina contour delineation is crucial for quantitative analyses of intravascular ultrasound (IVUS) images. However, the various artifacts in IVUS images pose substantial challenges for accurate delineation. Existing mask-based methods often produce anatomically implausible contours in artifact-affected images, while contour-based methods suffer from the over-smooth problem within the artifact regions. In this paper, we directly regress the contour pairs instead of mask-based segmentation. A coupled contour representation is adopted to learn a low-dimensional contour signature space, where the embedded anatomical prior enables the model to avoid producing unreasonable results. Further, a PIoU loss is proposed to capture the overall shape of the contour points and maximize the similarity between the regressed contours and manually delineated contours with various irregular shapes, alleviating the over-smooth problem. For the images with severe artifacts, a difficulty-aware training strategy is designed for contour regression, which gradually guides the model focus on hard samples and improves contour localization accuracy. We evaluate the proposed framework on a large IVUS dataset, consisting of 7204 frames from 185 pullbacks. The mean Dice similarity coefficients of the method for the lumen and external elastic lamina are 0.951 and 0.967, which significantly outperforms other state-of-the-art (SOTA) models. All regressed contours in the test images are anatomically plausible. On the public IVUS-2011 dataset, the proposed method attains comparable performance to the SOTA models with the highest processing speed at 100 fps. The code is available at https://github.com/SMU-MedicalVision/ContourRegression.