Cephalometric Landmark Detection (CLD), i.e. annotating interest points in lateral X-ray images, is the crucial first step of every orthodontic therapy. While CLD has immense potential for automation using Deep Learning methods, carefully crafted contemporary approaches using convolutional neural networks and heatmap prediction do not qualify for large-scale clinical application due to insufficient performance. We propose a novel approach using Vision Transformers (ViTs) with direct coordinate prediction, avoiding the memory-intensive heatmap prediction common in previous work. Through extensive ablation studies comparing our method against contemporary CNN architectures (ConvNext V2) and heatmap-based approaches (Segformer), we demonstrate that ViTs with coordinate prediction achieve superior performance with more than 2 mm improvement in mean radial error compared to state-of-the-art CLD methods. Our results show that while non-adapted CNN architectures perform poorly on the given task, contemporary approaches may be too tailored to specific datasets, failing to generalize to different and especially sparse datasets. We conclude that using general-purpose Vision Transformers with direct coordinate prediction shows great promise for future research on CLD and medical computer vision.

Microbeam Radiation Therapy (MRT) is an innovative radiotherapy modality which uses highly focused synchrotron-generated X-ray microbeams. Current pre-clinical research in MRT mostly rely on Monte Carlo (MC) simulations for dose estimation, which are highly accurate but computationally intensive. Recently, Deep Learning (DL) dose engines have been proved effective in generating fast and reliable dose distributions in different RT modalities. However, relatively few studies compare different models on the same task. This work aims to compare a Graph-Convolutional-Network-based DL model, developed in the context of Very High Energy Electron RT, to the Convolutional 3D U-Net that we recently implemented for MRT dose predictions. The two DL solutions are trained with 3D dose maps, generated with the MC-Toolkit Geant4, in rats used in MRT pre-clinical research. The models are evaluated against Geant4 simulations, used as ground truth, and are assessed in terms of Mean Absolute Error, Mean Relative Error, and a voxel-wise version of the γ-index. Also presented are specific comparisons of predictions in relevant tumor regions, tissues boundaries and air pockets. The two models are finally compared from the perspective of the execution time and size. This study finds that the two models achieve comparable overall performance. Main differences are found in their dosimetric accuracy within specific regions, such as air pockets, and their respective inference times. Consequently, the choice between models should be guided primarily by data structure and time constraints, favoring the graph-based method for its flexibility or the 3D U-Net for its faster execution.

Brain metastasis (BM) significantly affects the prognosis of non-small cell lung cancer (NSCLC) patients. Increasing evidence suggests that adipose tissue influences cancer progression and metastasis. This study aimed to develop a predictive nomogram integrating mediastinal fat area (MFA) and deep learning (DL)-derived tumor characteristics to stratify postoperative‌ BM risk in NSCLC patients. A retrospective cohort of 585 surgically resected NSCLC patients was analyzed. Preoperative computed tomography (CT) scans were utilized to quantify MFA using ImageJ software (radiologist-validated measurements). Concurrently, a DL algorithm extracted tumor radiomic features, generating a deep learning brain metastasis score (DLBMS). Multivariate logistic regression identified independent BM predictors, which were incorporated into a nomogram. Model performance was assessed via area under the receiver operating characteristic curve (AUC), calibration plots, integrated discrimination improvement (IDI), net reclassification improvement (NRI), and decision curve analysis (DCA). Multivariate analysis identified N stage, EGFR mutation status, MFA, and DLBMS as independent predictors of BM. The nomogram achieved superior discriminative capacity (AUC: 0.947 in the test set), significantly outperforming conventional models. MFA contributed substantially to predictive accuracy, with IDI and NRI values confirming its incremental utility (IDI: 0.123, <i>P</i> < 0.001; NRI: 0.386, <i>P</i> = 0.023). Calibration analysis demonstrated strong concordance between predicted and observed BM probabilities, while DCA confirmed clinical net benefit across risk thresholds. This DL-enhanced nomogram, incorporating MFA and tumor radiomics, represents a robust and clinically useful tool for preoperative prediction of postoperative BM risk in NSCLC. The integration of adipose tissue metrics with advanced imaging analytics advances personalized prognostic assessment in NSCLC patients. The online version contains supplementary material available at 10.1186/s12885-025-14466-5.

Early detection of breast cancer through mammography significantly improves survival rates. However, high false positive and false negative rates remain a challenge. Deep learning-based computer-aided diagnosis systems can assist in lesion detection, but their performance is often limited by the availability of labeled clinical data. This study systematically evaluated the effectiveness of transfer learning, image preprocessing techniques, and the latest You Only Look Once (YOLO) model (v9) for optimizing breast mass detection models on small proprietary datasets. We examined 133 mammography images containing masses and assessed various preprocessing strategies, including cropping and contrast enhancement. We further investigated the impact of transfer learning using the OPTIMAM Mammography Image Database (OMI-DB) compared with training on proprietary data. The performance of YOLOv9 was evaluated against YOLOv7 to determine improvements in detection accuracy. Pretraining on the OMI-DB dataset with cropped images significantly improved model performance, with YOLOv7 achieving a 13.9 % higher mean average precision (mAP) and 13.2 % higher F1-score compared to training only on proprietary data. Among the tested models and configurations, the best results were obtained using YOLOv9 pretrained OMI-DB and fine-tuned with cropped proprietary images, yielding an mAP of 73.3 % ± 16.7 % and an F1-score of 76.0 % ± 13.4 %, under this condition, YOLOv9 outperformed YOLOv7 by 8.1 % in mAP and 9.2 % in F1-score. This study provides a systematic evaluation of transfer learning and preprocessing techniques for breast mass detection in small datasets. Our results demonstrating that YOLOv9 with OMI-DB pretraining significantly enhances the performance of breast mass detection models while reducing training time, providing a valuable guideline for optimizing deep learning models in data-limited clinical applications.

This review explores the significant progress and applications of artificial intelligence (AI) in obstetrics and gynecological MRI, charting its development from foundational algorithmic techniques to deep learning strategies and advanced radiomics. This review features research published over the last few years that has used AI with MRI to identify specific conditions such as uterine leiomyosarcoma, endometrial cancer, cervical cancer, ovarian tumors, and placenta accreta. In addition, it covers studies on the application of AI for segmentation and quality improvement in obstetrics and gynecology MRI. The review also outlines the existing challenges and envisions future directions for AI research in this domain. The growing accessibility of extensive datasets across various institutions and the application of multiparametric MRI are significantly enhancing the accuracy and adaptability of AI. This progress has the potential to enable more accurate and efficient diagnosis, offering opportunities for personalized medicine in the field of obstetrics and gynecology.

This paper presents a novel motion feature guided diffusion model for unpaired video-to-video translation (MFD-V2V), designed to synthesize dynamic, high-contrast cine cardiac magnetic resonance (CMR) from lower-contrast, artifact-prone displacement encoding with stimulated echoes (DENSE) CMR sequences. To achieve this, we first introduce a Latent Temporal Multi-Attention (LTMA) registration network that effectively learns more accurate and consistent cardiac motions from cine CMR image videos. A multi-level motion feature guided diffusion model, equipped with a specialized Spatio-Temporal Motion Encoder (STME) to extract fine-grained motion conditioning, is then developed to improve synthesis quality and fidelity. We evaluate our method, MFD-V2V, on a comprehensive cardiac dataset, demonstrating superior performance over the state-of-the-art in both quantitative metrics and qualitative assessments. Furthermore, we show the benefits of our synthesized cine CMRs improving downstream clinical and analytical tasks, underscoring the broader impact of our approach. Our code is publicly available at https://github.com/SwaksharDeb/MFD-V2V.

Medical digitization has been intensively developed in the last decade, leading to paving the path for computer-aided medical diagnosis research. Thus, anomaly detection based on machine and deep learning techniques has been extensively employed in healthcare applications, such as medical imaging classification and monitoring of patients' vital signs. To effectively leverage digitized medical records for identifying challenges in healthcare, this manuscript presents a smart Clinical Decision Support System (CDSS) dedicated for medical multimodal data automated diagnosis. A smart healthcare system necessitating medical data management and decision-making is proposed. To deliver timely rapid diagnosis, thread-level parallelism (TLP) is utilized for parallel distribution of classification tasks on three edge computing devices, each employing an AI module for on-device AI classifications. In comparison to existing machine and deep learning classification techniques, the proposed multithreaded architecture realizes a hybrid (ML and DL) processing module on each edge node. In this context, the presented edge computing-based parallel architecture captures a high level of parallelism, tailored for dealing with multiple categories of medical records. The cluster of the proposed architecture encompasses three edge computing Raspberry Pi devices and an edge server. Furthermore, lightweight neural networks, such as MobileNet, EfficientNet, and ResNet18, are trained and optimized based on genetic algorithms to provide classification of brain tumor, pneumonia, and colon cancer. Model deployment was conducted based on Python programming, where PyCharm is run on the edge server whereas Thonny is installed on edge nodes. In terms of accuracy, the proposed GA-based optimized ResNet18 for pneumonia diagnosis achieves 93.59% predictive accuracy and reduces the classifier computation complexity by 33.59%, whereas an outstanding accuracy of 99.78% and 100% were achieved with EfficientNet-v2 for brain tumor and colon cancer prediction, respectively, while both models preserving a reduction of 25% in the model's classifier. More importantly, an inference speedup of 28.61% and 29.08% was obtained by implementing parallel 2 DL and 3 DL threads configurations compared to the sequential implementation, respectively. Thus, the proposed multimodal-multithreaded architecture offers promising prospects for comprehensive and accurate anomaly detection of patients' medical imaging and vital signs. To summarize, our proposed architecture contributes to the advancement of healthcare services, aiming to improve patient medical diagnosis and therapy outcomes.

PurposePrecise quantification of myocardial blood flow (MBF) and flow reserve (MFR) in 18F-flurpiridaz PET significantly relies on motion correction (MC). However, the manual frame-by-frame correction leads to significant inter-observer variability, time-consuming, and requires significant experience. We propose a deep learning (DL) framework for automatic MC of 18F-flurpiridaz PET. MethodsThe method employs a 3D ResNet based architecture that takes 3D PET volumes and outputs motion vectors. It was validated using 5-fold cross-validation on data from 32 sites of a Phase III clinical trial (NCT01347710). Manual corrections from two experienced operators served as ground truth, and data augmentation using simulated vectors enhanced training robustness. The study compared the DL approach to both manual and standard non-AI automatic MC methods, assessing agreement and diagnostic accuracy using minimal segmental MBF and MFR. ResultsThe area under the receiver operating characteristic curves (AUC) for significant CAD were comparable between DL-MC MBF, manual-MC MBF from Operators (AUC=0.897,0.892 and 0.889, respectively; p>0.05), standard non-AI automatic MC (AUC=0.877; p>0.05) and significantly higher than No-MC (AUC=0.835; p<0.05). Similar findings were observed with MFR. The 95% confidence limits for agreement with the operator were {+/-}0.49ml/g/min (mean difference = 0.00) for MFR and {+/-}0.24ml/g/min (mean difference = 0.00) for MBF. ConclusionDL-MC is significantly faster but diagnostically comparable to manual-MC. The quantitative results obtained with DL-MC for MBF and MFR are in excellent agreement with those manually corrected by experienced operators compared to standard non-AI automatic MC in patients undergoing 18F-flurpiridaz PET-MPI.