De-identification of medical images is a critical step to ensure privacy during data sharing in research and clinical settings. The initial step in this process involves detecting Protected Health Information (PHI), which can be found in image metadata or imprinted within image pixels. Despite the importance of such systems, there has been limited evaluation of existing AI-based solutions, creating barriers to the development of reliable and robust tools. In this study, we present an AI-based pipeline for PHI detection, comprising three key modules: text detection, text extraction, and text analysis. We benchmark three models-YOLOv11, EasyOCR, and GPT-4o- across different setups corresponding to these modules, evaluating their performance on two different datasets encompassing multiple imaging modalities and PHI categories. Our findings indicate that the optimal setup involves utilizing dedicated vision and language models for each module, which achieves a commendable balance in performance, latency, and cost associated with the usage of large language models (LLMs). Additionally, we show that the application of LLMs not only involves identifying PHI content but also enhances OCR tasks and facilitates an end-to-end PHI detection pipeline, showcasing promising outcomes through our analysis.

Magnetic resonance imaging (MRI) is the gold standard for diagnosing disc displacement in temporomandibular joint (TMJ) disorders, but its high cost and practical challenges limit its accessibility. This study aimed to develop a machine learning (ML) model that can predict TMJ disc displacement using only cone-beam computed tomography (CBCT)-based radiomics features without MRI. CBCT images of 247 mandibular condyles from 134 patients who also underwent MRI scans were analyzed. To conduct three experiments based on the classification of various patient groups, we trained two ML models, random forest (RF) and extreme gradient boosting (XGBoost). Experiment 1 classified the data into three groups: Normal, disc displacement with reduction (DDWR), and disc displacement without reduction (DDWOR). Experiment 2 classified Normal versus disc displacement group (DDWR and DDWOR), and Experiment 3 classified Normal and DDWR versus DDWOR group. The RF model showed higher performance than XGBoost across all three experiments, and in particular, Experiment 3, which differentiated DDWOR from other conditions, achieved the highest accuracy with an area under the receiver operating characteristic curve (AUC) values of 0.86 (RF) and 0.85 (XGBoost). Experiment 2 followed with AUC values of 0.76 (RF) and 0.75 (XGBoost), while Experiment 1, which classified all three groups, had the lowest accuracy of 0.63 (RF) and 0.59 (XGBoost). The RF model, utilizing radiomics features from CBCT images, demonstrated potential as an assistant tool for predicting DDWOR, which requires the most careful management.

Efficient processing of radiology reports for monitoring disease progression is crucial in oncology. Although large language models (LLMs) show promise in extracting structured information from medical reports, privacy concerns limit their clinical implementation. This study evaluates the feasibility and accuracy of two of the most recent Llama models for generating structured lymphoma progression reports from cross-sectional imaging data in a privacy-preserving, real-world clinical setting. This single-center, retrospective study included adult lymphoma patients who underwent cross-sectional imaging and treatment between July 2023 and July 2024. We established a chain-of-thought prompting strategy to leverage the locally deployed Llama-3.3-70B-Instruct and Llama-4-Scout-17B-16E-Instruct models to generate lymphoma disease progression reports across three iterations. Two radiologists independently scored nodal and extranodal involvement, as well as Lugano staging and treatment response classifications. For each LLM and task, we calculated the F1 score, accuracy, recall, precision, and specificity per label, as well as the case-weighted average with 95% confidence intervals (CIs). Both LLMs correctly implemented the template structure for all 65 patients included in this study. Llama-4-Scout-17B-16E-Instruct demonstrated significantly greater accuracy in extracting nodal and extranodal involvement information (nodal: 0.99 [95% CI = 0.98-0.99] vs. 0.97 [95% CI = 0.95-0.96], p < 0.001; extranodal: 0.99 [95% CI = 0.99-1.00] vs. 0.99 [95% CI = 0.98-0.99], p = 0.013). This difference was more pronounced when predicting Lugano stage and treatment response (stage: 0.85 [95% CI = 0.79-0.89] vs. 0.60 [95% CI = 0.53-0.67], p < 0.001; treatment response: 0.88 [95% CI = 0.83-0.92] vs. 0.65 [95% CI = 0.58-0.71], p < 0.001). Neither model produced hallucinations of newly involved nodal or extranodal sites. The highest relative error rates were found when interpreting the level of disease after treatment. In conclusion, privacy-preserving LLMs can effectively extract clinical information from lymphoma imaging reports. While they excel at data extraction, they are limited in their ability to generate new clinical inferences from the extracted information. Our findings suggest their potential utility in streamlining documentation and highlight areas requiring optimization before clinical implementation.

Radiography imaging protocols target on specific anatomical regions, resulting in highly consistent images with recurrent structural patterns across patients. Recent advances in medical anomaly detection have demonstrated the effectiveness of CNN- and transformer-based approaches. However, CNNs exhibit limitations in capturing long-range dependencies, while transformers suffer from quadratic computational complexity. In contrast, Mamba-based models, leveraging superior long-range modeling, structural feature extraction, and linear computational efficiency, have emerged as a promising alternative. To capitalize on the inherent structural regularity of medical images, this study introduces SP-Mamba, a spatial-perception Mamba framework for unsupervised medical anomaly detection. The window-sliding prototype learning and Circular-Hilbert scanning-based Mamba are introduced to better exploit consistent anatomical patterns and leverage spatial information for medical anomaly detection. Furthermore, we excavate the concentration and contrast characteristics of anomaly maps for improving anomaly detection. Extensive experiments on three diverse medical anomaly detection benchmarks confirm the proposed method's state-of-the-art performance, validating its efficacy and robustness. The code is available at https://github.com/Ray-RuiPan/SP-Mamba.

Apical periodontitis is a prevalent oral pathology that presents significant public health challenges. Despite advances in automated diagnostic systems across various medical fields, the development of Computer-Aided Diagnosis (CAD) applications for apical periodontitis is still constrained by the lack of a large-scale, high-quality annotated dataset. To address this issue, we release a large-scale panoramic radiograph benchmark called "PerioXrays", comprising 3,673 images and 5,662 meticulously annotated instances of apical periodontitis. To the best of our knowledge, this is the first benchmark dataset for automated apical periodontitis diagnosis. This paper further proposes a clinical-oriented apical periodontitis detection (PerioDet) paradigm, which jointly incorporates Background-Denoising Attention (BDA) and IoU-Dynamic Calibration (IDC) mechanisms to address the challenges posed by background noise and small targets in automated detection. Extensive experiments on the PerioXrays dataset demonstrate the superiority of PerioDet in advancing automated apical periodontitis detection. Additionally, a well-designed human-computer collaborative experiment underscores the clinical applicability of our method as an auxiliary diagnostic tool for professional dentists.

Precise 3D segmentation of paranasal sinuses is essential for accurate diagnosis and treatment. This study aimed to develop a fully automated segmentation algorithm for the paranasal sinuses using the nnU-Net v2 architecture. The nnU-Net v2-based segmentation algorithm was developed using Python 3.6.1 and the PyTorch library, and its performance was evaluated on a dataset of 97 cone-beam computed tomography (CBCT) scans. Ground truth annotations were manually generated by expert radiologists using the 3D Slicer software, employing a polygonal labeling technique across sagittal, coronal, and axial planes. Model performance was assessed using several quantitative metrics, including accuracy, Dice Coefficient (DC), sensitivity, precision, Jaccard Index, Area Under the Curve (AUC), and 95% Hausdorff Distance (95% HD). The nnU-Net v2-based algorithm demonstrated high segmentation performance across all paranasal sinuses. Dice Coefficient (DC) values were 0.94 for the frontal, 0.95 for the sphenoid, 0.97 for the maxillary, and 0.88 for the ethmoid sinuses. Accuracy scores exceeded 99% for all sinuses. The 95% Hausdorff Distance (95% HD) values were 0.51 mm for both the frontal and maxillary sinuses, 0.85 mm for the sphenoid sinus, and 1.17 mm for the ethmoid sinus. Jaccard indices were 0.90, 0.91, 0.94, and 0.80, respectively. This study highlights the high accuracy and precision of the nnU-Net v2-based CNN model in the fully automated segmentation of paranasal sinuses from CBCT images. The results suggest that the proposed model can significantly contribute to clinical decision-making processes, facilitating diagnostic and therapeutic procedures.

Purpose: Accurate tumor segmentation is vital for adaptive radiation therapy (ART) but remains time-consuming and user-dependent. Segment Anything Model 2 (SAM2) shows promise for prompt-based segmentation but struggles with tumor accuracy. We propose prior knowledge-based augmentation strategies to enhance SAM2 for ART. Methods: Two strategies were introduced to improve SAM2: (1) using prior MR images and annotations as contextual inputs, and (2) improving prompt robustness via random bounding box expansion and mask erosion/dilation. The resulting model, SAM2-Aug, was fine-tuned and tested on the One-Seq-Liver dataset (115 MRIs from 31 liver cancer patients), and evaluated without retraining on Mix-Seq-Abdomen (88 MRIs, 28 patients) and Mix-Seq-Brain (86 MRIs, 37 patients). Results: SAM2-Aug outperformed convolutional, transformer-based, and prompt-driven models across all datasets, achieving Dice scores of 0.86(liver), 0.89(abdomen), and 0.90(brain). It demonstrated strong generalization across tumor types and imaging sequences, with improved performance in boundary-sensitive metrics. Conclusions: Incorporating prior images and enhancing prompt diversity significantly boosts segmentation accuracy and generalizability. SAM2-Aug offers a robust, efficient solution for tumor segmentation in ART. Code and models will be released at https://github.com/apple1986/SAM2-Aug.

This study aims to explore a novel approach for predicting cervical lymph node metastasis (CLNM) and pathological subtypes in oral squamous cell carcinoma (OSCC) by comparing deep learning (DL) and habitat analysis models based on contrast-enhanced CT (CECT). A retrospective analysis was conducted using CECT images from patients diagnosed with OSCC via paraffin pathology at the Second Affiliated Hospital of Dalian Medical University. All patients underwent primary tumor resection and cervical lymph node dissection, with a total of 132 cases included. A DL model was developed by analysing regions of interest (ROIs) in the CECT images using a convolutional neural network (CNN). For habitat analysis, the ROI images were segmented into 3 regions using K-means clustering, and features were selected through a fully connected neural network (FCNN) to build the model. A separate clinical model was constructed based on nine clinical features, including age, gender, and tumor location. Using LNM and pathological subtypes as endpoints, the predictive performance of the clinical model, DL model, habitat analysis model, and a combined clinical + habitat model was evaluated using confusion matrices and receiver operating characteristic (ROC) curves. For LNM prediction, the combined clinical + habitat model achieved an area under the ROC curve (AUC) of 0.97. For pathological subtype prediction, the AUC was 0.96. The DL model yielded an AUC of 0.83 for LNM prediction and 0.91 for pathological subtype classification. The clinical model alone achieved an AUC of 0.94 for predicting LNM. The integrated habitat-clinical model demonstrates improved predictive performance. Combining habitat analysis with clinical features offers a promising approach for the prediction of oral cancer. The habitat-clinical integrated model may assist clinicians in performing accurate preoperative prognostic assessments in patients with oral cancer.

Medical image segmentation is a complex and challenging task, which aims to accurately segment various structures or abnormal regions in medical images. However, obtaining accurate segmentation results is difficult because of the great uncertainty in the shape, location, and scale of the target region. To address these challenges, we propose a higher-order spatial interaction framework with dual cross global efficient attention (DGEAHorNet), which employs a neural network architecture based on recursive gate convolution to adequately extract multi-scale contextual information from images. Specifically, a Dual Cross-Attentions (DCA) is added to the skip connection that can effectively blend multi-stage encoder features and narrow the semantic gap. In the bottleneck stage, global channel spatial attention module (GCSAM) is used to extract image global information. To obtain better feature representation, we feed the output from the GCSAM into the multi-branch dense layer (SENetV2) for excitation. Furthermore, we adopt Depthwise Over-parameterized Convolutional Layer (DO-Conv) in order to replace the common convolutional layer in the input and output part of our network, then add Efficient Attention (EA) to diminish computational complexity and enhance our model's performance. For evaluating the effectiveness of our proposed DGEAHorNet, we conduct comprehensive experiments on four publicly-available datasets, and achieving 0.9320, 0.9337, 0.9312 and 0.7799 in Dice similarity coefficient on ISIC2018, ISIC2017, CVC-ClinicDB and HRF respectively. Our results show that DGEAHorNet has better performance compared with advanced methods. The code is publicly available at https://github.com/penghaixin/mymodel .

Dual-energy X-ray Computed Tomography (DECT) constitutes an advanced technology which enables automatic decomposition of materials in clinical images without manual segmentation using the dependency of the X-ray linear attenuation with energy. However, most methods perform material decomposition in the image domain as a post-processing step after reconstruction but this procedure does not account for the beam-hardening effect and it results in sub-optimal results. In this work, we propose a deep learning procedure called Dual-Energy Decomposition Model-based Diffusion (DEcomp-MoD) for quantitative material decomposition which directly converts the DECT projection data into material images. The algorithm is based on incorporating the knowledge of the spectral DECT model into the deep learning training loss and combining a score-based denoising diffusion learned prior in the material image domain. Importantly the inference optimization loss takes as inputs directly the sinogram and converts to material images through a model-based conditional diffusion model which guarantees consistency of the results. We evaluate the performance with both quantitative and qualitative estimation of the proposed DEcomp-MoD method on synthetic DECT sinograms from the low-dose AAPM dataset. Finally, we show that DEcomp-MoD outperform state-of-the-art unsupervised score-based model and supervised deep learning networks, with the potential to be deployed for clinical diagnosis.