Breast cancer is a leading cause of cancer-related deaths among women. Advances in early diagnosis and treatment, particularly through screening mammography, have reduced mortality rates by enabling the detection of small tumors. Recently, artificial intelligence (AI) and advanced computer vision models have further improved breast cancer detection and diagnosis. In this research, we have developed a novel model for detecting breast cancer in mammography images by extracting rich and suitable features. Our model utilizes the Neighborhood Attention Transformer, which enhances local feature processing by focusing on neighborhood attention alongside global and long-range features. This is crucial for analyzing masses within and at the boundaries. Additionally, we incorporate the Shearlet Transform to enhance feature extraction by capturing frequency-domain features, essential for precise edge and texture analysis in mammographic images. The Shearlet Transform's ability to manage anisotropic features and its strong localization in both spatial and frequency domains make it particularly effective. Denoising is another key aspect, as mammograms often contain noise from imaging conditions and devices. To address this, our model applies Shearlet-based adaptive shrinkage denoising, significantly improving feature extraction. By combining the energy of Shearlet subbands with features from previous techniques, our model simplifies feature representation, highlights key patterns, and remains robust to noise and transformations. Our proposed model has achieved impressive results on the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) dataset, with F1, Area Under the Curve (AUC), and Cohen's Kappa scores of 76.8 %, 84.5 %, and 50.9 %, respectively, outperforming other models.

To evaluate the performance of various computed tomography angiography (CTA) reconstruction methods in assessing coronary plaque composition using near-infrared spectroscopy-intravascular ultrasound (NIRS-IVUS) as the reference standard. Fifteen patients with chronic coronary syndrome underwent CTA and 3-vessel NIRS-IVUS imaging. CTA datasets were reconstructed using a medium-smooth b40f kernel with two slice thicknesses, 0.75 ​mm and 0.50 ​mm, and three strengths of advanced model-based iterative reconstruction (ADMIRE). Plaque components on CTA were classified using fixed and adaptive Hounsfield unit (HU) thresholds while NIRS-IVUS classification employed a deep learning method validated against histology. Matched CTA and NIRS-IVUS images were analyzed to quantify fibrotic tissue (FT), necrotic core (NC), and calcific (Ca) volumes and areas at both segment- and lesion-level. The intraclass correlation coefficient (ICC) and Surface Under the Cumulative Ranking Curve (SUCRA) scores were used to determine the best-performing reconstruction approach. Fifty vessels were included in the final analysis. CTA showed weak correlation with NIRS-IVUS for FT (ICC <0.43), good correlation for Ca (ICC 0.42-0.83), and poor correlation for NC, except when using reconstruction approach ADMIRE 2, 0.50 ​mm slice thickness, and fixed HU cutoffs, which demonstrated moderate correlation for NC (segment-level ICC ​= ​0.67; lesion-level ICC ​= ​0.61). This approach ranked highest on SUCRA plots. CTA reconstruction parameters influence plaque composition analysis. The combination of an intermediate-strength IR, thin slice thickness, and fixed HU cutoffs yields the most accurate tissue characterization compared to NIRS-IVUS as the reference standard.

To establish normative three-dimensional airway measurements in patients without dentofacial deformities (DDFs) or obstructive sleep apnea (OSA), and to identify anatomical and epidemiological factors associated with airway volume. This retrospective cross-sectional study analyzed 200 CT scans from patients aged 18-80 years, with no diagnosis of DDF, OSA, or craniofacial syndromes. Scans were processed using artificial intelligence software (NEMOFAB) for automatic segmentation and volumetric analysis. Variables assessed included age, sex, neck circumference, and craniofacial linear distances (Menton-Hyoid, Menton-3rd Vertebrae, PNS-Hyoid, Soft Palate-Hyoid). Airway volume and Minimum Axial Area (MAA) were measured and compared using ANOVA. The mean airway volume was 24,724.8mm³. Younger individuals exhibited greater airway volumes, especially among males. Patients with a neck circumference <40cm had a 28.04% reduction in airway volume. Longer PNS-Hyoid, SPH, and M3V distances were positively associated with increased airway volume, while Menton-Hyoid showed minimal impact. A low MAA (<110mm²) correlated with a significant volume decrease. Key predictors identified were age, neck circumference, PNS-Hyoid, SPH, and M3V distances. In patients without DDF or OSA, airway volume is significantly influenced by demographic and anatomical variables. These normative data provide a baseline for comparison in orthognathic surgical planning and respiratory risk assessment. Understanding normal airway morphology and its anatomical determinants enhances screening for patients at risk of airway compromise and may guide individualized treatment strategies in oral and maxillofacial surgery.

Three-dimensional (3D) landmark detection is essential for assessing craniofacial growth and planning surgeries, such as orthodontic, orthognathic, traumatic, and plastic procedures. This study aimed to develop an automatic 3D landmarking model for oral and maxillofacial regions and to validate its accuracy, robustness and generalizability in both spiral computed tomography (SCT, 41 landmarks) and cone-beam computed tomography (CBCT, 14 landmarks) scans. The model was implemented using an optimized lightweight 3D U-Net network architecture. Its accuracy, robustness and generalizability were thoroughly evaluated and validated through a multicenter retrospective diagnostic study. The model was trained and tested on a data set of 480 SCT and 240 CBCT cases. An additional inference on a different data set of 320 SCT and 150 CBCT cases was performed. Mean radial error (MRE) and success detection rate within 2-, 3-, and 4-mm error thresholds were measured as the primary evaluation metrics. Error analyses for landmark detection along each coordinate axis were performed. Consistency tests among observers were conducted. The average MRE for both SCT and CBCT was consistently below 1.3 mm and, notably, below 1.4 mm in complex conditions, such as malocclusion, missing dental landmarks, and the presence of metal artifacts. No significant differences in MRE and SDR at 2-4 mm were observed between external and internal SCT and CBCT sets. SCT bone landmarks were more precise than dental ones, with no difference between bone/soft tissue and dental/soft tissue. CBCT dental landmarks exhibited greater precision compared to bone landmarks. A detailed error analysis across the coordinate axes showed that the coronal axis had the highest error rates. The implementation of this model significantly improved the landmarking proficiency of senior and junior specialists by 15.9% and 28.9%, respectively, while also achieving a 6-9.5-fold acceleration in GUI interaction time. This study shows that the AI-driven model delivers high-precision 3D localization of oral and maxillofacial landmarks, even in complex scenarios. The model demonstrates potential as a promising computer-aided tool to assist specialists in conducting accurate and efficient localization analyses; however, its robustness and generalizability require prospective clinical validation to ensure utility across varied experience levels.

Recurrent lumbar disc herniation (RLDH) is a significant challenge following lumbar discectomy, with recurrence rates of 5%-15%. Established risk factors include male gender, diabetes mellitus, smoking, and obesity, but the role of paraspinal muscles in recurrence is unclear. This study was conducted to identify key risk factors for RLDH, including the volume of paraspinal muscles with machine learning. We used data from 126 patients who underwent lumbar discectomy between January 2003 and September 2023 and had follow-up outpatient visits for more than 6 months at a single institution. Variables selected for the model, comprising demographic and clinical variables, medical history, LDH operation-related variables, and MRI measurements for RLDH. Based on clinical symptoms and radiologic results, the patients were classified into RLDH and non-RLDH groups, and RLDH was defined at the same surgical level on follow-up MRI. Totally, 38 patients were included in the RLDH group and 88 in the non-RLDH group. The volume of quadratus lumborum was identified as a risk factor for RLDH (odds ratio 7.894; P=0.001). Among the five different ML algorithms, XGBoost achieved the best result with an accuracy of 0.794 and area under the curve (AUC) of 0.811. In terms of SHAP value analysis, the weight, volume of quadratus lumborum, psoas major, and vertebra were key features for predicting RLDH. The prediction model would be of great assistance for surgeons to make surgical decisions or establish observation intervals.

Parkinson's Disease (PD) is a progressive neurodegenerative disorder primarily characterized by the gradual loss of dopamine-producing neurons in the brain's substantia nigra. The hallmark motor symptoms of PD include tremors, bradykinesia (slowness of movement), rigidity (muscle stiffness), and postural instability. However, the disease also manifests with significant non-motor symptoms such as cognitive decline, mood disorders, sleep disturbances, and autonomic dysfunction, which further complicate the clinical images. Accurate and early diagnosis of PD is challenging due to the subtlety and gradual onset of symptoms, as well as the overlap with other neurodegenerative disorders. Traditional diagnostic methods rely heavily on clinical evaluations and motor symptom assessments, which can be subjective and not detect early or asymptomatic stages of the disease. To overcome these challenges, this work aims to propose a novel Feature-level Fusion-enabled Parkinson's Disease Detection (FLF-PDD) system, integrating an Improved Bidirectional- Gated Recurrent Unit (Bi-GRU) architecture. This model progresses through several stages: preprocessing, where an Enhanced Gaussian Filtering technique reduces image noise; feature extraction, employing methods such as Enhanced Pyramid Histograms of Oriented Gradients (PHOG), Multi Texton, LGXP, and color analysis; feature-level fusion, utilizing Principal Component Analysis (PCA) and tanh normalization to combine extracted features from various MRI orientations; and disease detection, facilitated by the trained Improved Bi-GRU model on fused features to accurately diagnose PD symptoms. The FLF-PDD model undergoes rigorous evaluation to enhance diagnostic accuracy and deepen the understanding of PD progression.

The urosepsis after percutaneous nephrolithotomy (PCNL) is a critical health risk necessitating prompt medical identification and intervention. Nevertheless, a deficiency exists in the availability of a tool for precise and timely predictive analysis. The purpose is to establish a machine learning (ML) model using radiomic features and clinical data to predict urosepsis following PCNL. This study retrospectively included 401 patients with kidney stones from two centers who underwent PCNL. To enhance the dataset's equilibrium, the synthetic minority over-sampling technique for regression with Gaussian noise (SMOGN) was used to resample the training set. The screening of radiomics features and the construction of radiomics scores were completed by applying the Absolute Shrinkage Selection Operator (LASSO). Subsequently, the critical clinical indicators for urosepsis were pinpointed through the application of a multivariate logistic regression. The performance of seven ML algorithms was compared for the combined dataset that incorporated clinical variables and radiomics scores. The efficacy of these models was assessed through the implementation of a fivefold cross-validation process. Ultimately, the Shapley Additive exPlanations (SHAP) methodology was utilized to provide a visual and interpretative analysis of the optimal model. Among 401 patients, 30 cases (7.48%) were diagnosed with urosepsis. The radiomics score, established by 13 radiomics features, was combined with six important clinical features (including urine nitrite positivity, stone volume, mean intrarenal pressures, urine white blood cells, and operation time) to construct a combined dataset. Comparative analysis of seven machine learning (ML) models revealed that CatBoost demonstrated superior predictive performance. The model achieved area under the receiver operating characteristic curve (AUC-ROC) values of 0.88, 0.94, and 0.89 on the training, internal test, and external validation sets, respectively. Corresponding area under the precision-recall curve (AUC-PR) values were 0.92, 0.75, and 0.63. The SHAP value method identifies key features influencing prediction outcomes, with the radiomics score and urine nitrite positivity being the top contributors to the model. We deployed the optimal prediction model to a web for clinical application ( https://predictive-model-for-urosepsis.streamlit.app/ ). This study constructed a predictive model that incorporates clinical risk characteristics and radiomics scores to assess the risk of urosepsis after PCNL, with SHAP visualization for clinical physicians to formulate evaluation strategies.

Neonatal acute bilirubin encephalopathy (ABE) severely endangers the neonatal health. However, early clinical symptoms of ABE are nonspecific, often leading to missed diagnoses. The current study endeavors to establish a computer-assisted integrated model for clinical assessment and diagnosis of ABE. Diagnostic data from the ABE group and the hyperbilirubinemia without concurrent ABE (non-ABE) group were retrospectively analyzed. Patients were divided into a pre-training cohort, a training cohort, and two test cohorts. The training cohort and test cohort 1 were used to train and test a deep learning (DL) model integrating multimodal, self-supervised, and multi-instance learning. Test cohort 2 was used to compare the DL model with the radiologists. A total of 1048 magnetic resonance images from 262 patients were analyzed. The accuracy of the DL model and the area under the curve were 86.3% and 91.2% and 91.1% and 89.3% in test cohorts 1 and 2, respectively. This study integrated clinical and radiological data into DL models to accurately diagnose ABE, close to the proficiency level of senior radiologists. It provides a convenient, low-cost evaluation model for patient management decisions and physician diagnoses.

To quantitatively measure the volume of white matter hyperintensities (WMHs) in different parts of the brain in patients with different types of cognitive function and analyze the relationship between WMH volume and cognitive function to obtain a threshold WMH volume for the early detection and clinical assessment of cognitive dysfunction. The clinical data and magnetic resonance imaging (MRI) data of patients with WMHs indicated by cranial MR in the Department of General Medicine of Shandong Provincial Third Hospital were collected. The FLAIR sequence images of the patients were subsequently analyzed with computer automated detection technology. Through deep learning-based 3D reconstruction, the specific volumes of the patients' WMHs were obtained. Patients were divided into three groups according to the Fazekas scale score: Fazekas score 1, Fazekas score 2, and Fazekas score 3. The WMH volumes within each group were subsequently compared, and the correlations between the WMH volumes of the patients in each group and their Montreal Cognitive Assessment (MoCA) scores, Trail Making Test A (TMT-A) scores, Trail Making Test B (TMT-B) scores, age, duration of hypertension, duration of diabetes, basic information, etc., were analyzed. The patients were subsequently divided into a normal group (MoCA > 25) and a mild cognitive impairment group (18 < MoCA ≤ 25) on the basis of their MoCA scores. The WMH volumes in each group were then calculated separately. The cutoff values of the WMH volume for differentiating between the normal group and mild cognitive impairment group were obtained through receiver operating characteristic (ROC) curve analysis. The MoCA scores significantly differed among the three Fazekas score groups (r = - 0.5716, P < 0.0001). There were also statistically significant differences in the total volume of WMHs among the three groups (r = 0.7527, P < 0.0001). WMH volume was positively correlated with the TMT-A and TMT-B scores (r = 0.2345,  P< 0.05; r = 0.2404, P < 0.05) but negatively correlated with the MoCA score (r = - 0.4789, P < 0.0001). Moreover, WMH volume was positively associated with the duration of hypertension (F = 4.743, P < 0.05) but not with the duration of diabetes (F = 1.431, P = 0.2456). The cutoff value of WMH volume between the normal group and mild cognitive impairment group was 15.474900; at this value, the sensitivity of the WMH volume in discriminating the two groups was 0.808, and the specificity was 0.556. Automated detection technology can successfully be used to obtain the volume of WMHs in different parts of patients' brains. Since WMH volume is correlated with cognitive function scores, we can use MRI to identify and assess individuals who show potential early signs of cognitive dysfunction and administer early interventions. These findings provide potential preventive and therapeutic targets for the clinical diagnosis and treatment of cognitive dysfunction.

Accurate segmentation of gastric cavities from ultrasound images remains a challenging task due to the presence of ultrasound shadow and varying anatomical structures. To address these challenges, we collected a Gastric Ultrasound Image (GUSI) dataset using transabdominal techniques, after administering an echoic cellulose-based gastric ultrasound contrast agent (TUS-OCCA), and annotated the gastric cavity regions. We propose a model called Shadow Adaptive Tracing U-net (SATU-net) for gastric cavity segmentation on the GUSI dataset. SATU-net is specifically designed for gastric cavity segmentation in ultrasound images. The method introduces an Adaptive Shadow Tracing Module (ASTM), Shadow Separation Module (SSM), and an affine transformation mechanism to mitigate the impact of ultrasound shadow. The affine transformation aligns ultrasound image regions to reduce geometric distortion, while the ASTM dynamically tracks and compensates for ultrasound shadow, and the SSM extracts the shadow separation image. Extensive experiments on the gastric ultrasound dataset demonstrate that SATU-net achieves superior segmentation performance compared to several state-of-the-art deep learning methods, with an IoU improvement of 2.26% over the second-best competitor. Further robustness analysis and limited external validation provide preliminary evidence that SATU-net generalizes across diverse clinical scenarios. Our method provides a robust solution for ultrasound image segmentation and can be extended to other medical imaging tasks. Additionally, the ASTM module can be flexibly applied to existing network frameworks.