Current Medical Imaging - Current Issue

The objective is to develop and compare risk prediction models for Iodine Contrast Media (ICM)-related Acute Adverse Reactions (AAR) in patients without a prior history of such reactions, and to construct a nomogram based on the superior model.

546 patients without a history of ICM-related AAR who underwent ICM administration during CT contrast-enhanced scan were retrospectively enrolled, and divided into training (n=234), test (n=101), and external validation (n=211) sets. Clinical, medication information, and environmental factors were collected. Features were selected by univariate logistical analysis and least absolute shrinkage and selection operator, and four Machine Learning (ML) models, including Logistic Regression (LR), decision tree, k-nearst neighbors and linear support vector classification were used to construct ICM-related AAR risk prediction models were developed and evaluated using AUC, accuracy and F1 score. A nomogram was constructed based on the superior model.

History of ICM exposure and allergy due to other factors, hypertension, type of ICMs, ICM dose, oral metformin, hyperglycaemia, and glomerular filtration rate were selected for modeling (all p < 0.05). The LR model demonstrated superior performance, with AUCs of 0.894 (test set) and 0.814 (external validation), and was used to construct a clinically applicable nomogram.

The LR-based model effectively predicts ICM-related AAR risk using readily available clinical variables. It offers a practical tool for identifying high-risk patients prior to ICM administration, facilitating preventive measures.

LR can predict the risk of ICM-related AAR well in patients without a history of ICM-related AAR, and the corresponding nomogram is provided.

Construct and compare multiple machine learning models to predict lymph node (LN) metastasis in cervical cancer, utilizing radiomic features extracted from preoperative multi-parametric magnetic resonance imaging (MRI).

This study retrospectively enrolled 407 patients with cervical cancer who were randomly divided into a training cohort (n=284) and a validation cohort (n=123). A total of 4065 radiomic features were extracted from the tumor regions of interest on contrast-enhanced T1-weighted imaging, T2-weighted imaging, and diffusion-weighted imaging for each patient. The Mann-Whitney U test, Spearman correlation analysis, and selection operator Cox regression analysis were employed for radiomic feature selection. The relationship between MRI radiomic features and LN status was analyzed using five machine-learning algorithms. Model performance was evaluated by measuring the area under the receiver-operating characteristic curve (AUC) and accuracy (ACC). Moreover, Kaplan–Meier analysis was used to validate the prognostic value of selected clinical and radiomic characteristics.

LN metastasis was pathologically detected in 24.3% (99/407) of patients. Following a three-step feature selection, 18 radiomic features were employed for model construction. The XGBoost model exhibited superior performance compared to other models, achieving an AUC, accuracy, sensitivity, specificity, and F1 score of 0.9268, 0.8969, 0.7419, 0.9891, and 0.8364, respectively, on the validation set. Additionally, Kaplan−Meier curves indicated a significant correlation between radiomic scores and progression-free survival in cervical cancer patients (p < 0.05).

Among the machine learning models, XGBoost demonstrated the best predictive ability for LN metastasis and showed prognostic value through its radiomic score, highlighting its clinical potential.

Machine learning-based multi-parametric MRI radiomic analysis demonstrated promising performance in the preoperative prediction of LN metastasis and clinical prognosis in cervical cancer.

Microsurgical treatment of posterior circulation aneurysms remains challenging due to their deep location, complex anatomical exposure, and close proximity to critical neurovascular structures. Ensuring adequate collateral circulation is paramount for preventing ischemic complications. Indocyanine Green (ICG) and Fluorescein Video Angiography (FL-VAG) have emerged as effective intraoperative tools for assessing cerebral perfusion and guiding surgical decision-making.

We report the case of a 29-year-old male presenting with a thunderclap headache, nausea, and vomiting, subsequently diagnosed with a fusiform aneurysm at the P2-P3 junction of the left posterior cerebral artery. The patient underwent a subtemporal approach with partial posterior petrosectomy for aneurysm clipping and remodeling. Initially, an STA-P3 and PITA-P3 bypass were considered; however, intraoperative ICG and FL-VAG confirmed sufficient retrograde collateral flow, allowing the bypass procedure to be avoided. Postoperative imaging demonstrated patent circulation in the occipitotemporal region without ischemic compromise.

This case highlights the crucial role of intraoperative fluorescence imaging in refining surgical strategies for complex aneurysm clipping. ICG and FL-VAG enhance surgical precision by providing real-time perfusion assessment, reducing the need for additional vascular interventions, and improving patient outcomes.

To address insufficient segmentation accuracy in athletes' soft tissue injury analysis, this study proposes an enhanced Swin-Unet model with multi-scale feature fusion via the FPN (Feature Pyramid Network) and an adaptive window selection mechanism for dynamic receptive field adjustment.

A weighted hybrid loss function integrating Dice Loss, Cross-Entropy Loss, and boundary auxiliary loss optimizes segmentation precision and boundary recognition.

Evaluated on the OAI-ZIB dataset using 10-fold cross-validation, the model achieves a DSC (Dice Similarity Coefficient) of 0.978, outperforming baseline Swin-Unet and mainstream architectures. Superior performance is demonstrated in IoU (Intersection over Union) (0.968) and boundary Hausdorff distance (3.21), alongside significantly reduced diagnosis time (6.0 minutes vs. 16.8 minutes manually).

This framework enhances real-time medical imaging analysis for athlete injuries, offering improved accuracy, efficiency, and clinical utility in soft tissue segmentation tasks.

To evaluate the correlation between the effective cross-sectional area (eCSA) of the rotator cuff muscle measured using Dixon MRI and the outcomes of the Biodex Isokinetic Test.

The cross-sectional area (CSA) of the subscapularis (SSc), supraspinatus (SST), and infraspinatus+teres minor (ISTM) muscles of 87 patients who had undergone shoulder MRI and Biodex Isokinetic Test were measured in the oblique sagittal Y-view. The eCSA was calculated by multiplying the CSA by (1-fat fraction). Eight shoulder movements (FL60, EX60, FL180, EX180, ER60, IR60, ER180, and IR180) each assessed using four parameters (peak torque [PT], peak torque/body weight, torque at 30° [TQ30], and total work) were recorded on Biodex. Pearson correlation coefficients were calculated between eCSA and Biodex outcomes. Univariate regression analyses were conducted to identify the factors influencing the Biodex results. General linear models were applied to confirm the correlations between the eCSA and 32 Biodex parameters after adjusting for these factors.

The eCSA of the SSc, SST, and ISTM exhibited significant correlations with TQ30 at IR180 (r=0.549) and FL60 (r=0.522), PT at ER60 (r=0.656) and EX60 (r=0.575), and PT at ER60 (r=0.674) and FL180 (r=0.626), respectively. Age, sex, SST, and SSc tears were identified as factors influencing the Biodex results. FL60TQ30, EX60PT, and ER60PT exhibited significant associations with the eCSA of SSc, SST, and ISTM, respectively, after adjusting for these factors.

eCSA may be a useful quantitative imaging marker for assessing the function of the rotator cuff muscle. FL60TQ30, EX60PT, and ER60PT are useful Biodex indices for SSc, SST, and ISTM, respectively.

Liver fibrosis is a key pathological process that can progress to cirrhosis and liver failure. Although magnetic resonance elastography (MRE) is an established noninvasive method for fibrosis staging, its clinical application is limited by hardware dependence. The diagnostic utility of diffusion-weighted imaging-based virtual MRE (vMRE) and B1-corrected T1 mapping in liver fibrosis assessment remains to be further investigated.

Forty rabbits were included in the final analysis: CCl4-induced fibrosis (n=33) and control (n=7). Following Gd-EOB-DTPA administration, DWI and T1 mapping sequences were executed at 5 and 10 minutes. Diagnostic efficacy and correlations of vMRE and T1 mapping in a rabbit liver fibrosis model were evaluated.

Rabbits were classified into three groups: Control (n=7), Nonadvanced fibrosis (F1-F2, n=20), and Advanced fibrosis (F3-F4, n=13). The AUC values for T1post_5min, T1post_10min, rΔT1_10min, and μdiff in distinguishing controls from nonadvanced and advanced fibrosis groups were (0.78, 0.82, 0.71), (0.82, 0.85, 0.77), and (0.62, 0.69, 0.74), respectively, with μdiff showing (0.90, 0.93, 0.66). A significant positive correlation existed between μdiff and liver fibrosis grade (r=0.534, p<0.001).

μdiff correlated well with fibrosis severity and effectively identified fibrotic livers, but showed limited ability to distinguish fibrosis stages, likely due to overlapping tissue stiffness. B1-corrected T1 mapping offered complementary functional information, with the 10-minute post-contrast time point providing the best staging performance, thereby enhancing the overall diagnostic value.

Gd-EOB-DTPA-enhanced T1 mapping and DWI-based vMRE provide substantial noninvasive assessment of liver fibrosis.

Intracranial hemorrhage (IH) causes dementia and Alzheimer’s disease in the later stages. Until now, the accurate, early detection of IH, its prognosis, and therapeutic interventions have been a challenging task. Objective: A Multimodal Joint Fusion Sentiment Analysis (MJFSA) framework is proposed for the early detection and classification of IH, as well as sentiment analysis to support prognosis and therapeutic report generation.

MJFSA integrates radiological images and the radiological clinical narrative reports (RCNRs). In the proposed MJFSA model, MRI brain images are enhanced using the modified Contrast Limited Adaptive Histogram Equalization (M-CLAHE) algorithm. Enhanced images are processed with the proposed Tuned Temporal-GAN (Tuned-T-GAN) algorithm to generate temporal images. RCNRs are generated for temporal images using the Microsoft-Phi2 language model. Temporal images are processed with the Tuned-Vision Image Transformer (T-ViT) model to extract image features. On the other hand, the Bio-Bidirectional Encoder Representation Transformer (Bio-BERT) processes the RCNR texts for text feature extraction. Temporal image and RCNR text features are used for IH classification, such as intracerebral hemorrhage (ICH), epidural hemorrhage (EDH), subdural hemorrhage (SDH), and intraventricular hemorrhage (IVH), resulting in sentiment analysis for prognosis and therapeutic reports.

The MJFSA model has achieved an accuracy of 96.5% in prognosis sentiment analysis and 94.5% in therapeutic sentiment analysis.

The Multimodal Joint Fusion Sentiment Analysis (MJFSA) framework detects IH and classifies it using sentiment analysis for prognosis and therapeutic report generation.

The MJFSA model’s prognosis and therapeutic sentiment analysis report aims to support the early identification and management of risk factors associated with dementia and Alzheimer’s disease.

Artificial intelligence has significantly enhanced disease diagnosis in healthcare, particularly through Deep Learning (DL) and Federated Learning (FL) approaches. These technologies have shown promise in detecting ocular diseases using medical imaging while addressing challenges related to data privacy and security. FL enables collaborative learning without sharing sensitive medical data, making it an attractive solution for healthcare applications. This systematic review aims to analyze the advancements in AI-driven ocular disease detection, with a particular focus on FL-based approaches. The article evaluates the evolution, methodologies, challenges, and effectiveness of FL in enhancing diagnostic accuracy while ensuring data confidentiality.

The systematic review followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) framework to ensure transparency and reliability. Research articles published between 2017 and 2024 were identified using academic databases, including Web of Science, Scopus, IEEE Xplore, and PubMed. Studies focusing on DL and FL models for detecting ocular diseases were selected based on predefined inclusion and exclusion criteria. A comparative analysis of the methodologies, architectures, datasets, and performance metrics of different FL models has been presented.

The findings indicated that FL preserves data privacy while achieving diagnostic performance comparable to traditional centralized AI models. Various FL models, including FedAvg and FedProx, have been implemented for ocular disease detection, with high accuracy and efficiency. However, challenges, such as data heterogeneity, communication efficiency, and model convergence, persist.

FL represents a promising approach for ocular disease detection, balancing diagnostic accuracy with data privacy. Future research may focus on optimizing FL frameworks for improving scalability, communication efficiency, and integrating advanced privacy-preserving techniques.

This article presents a case of a patient with a broad ligament perivascular epithelioid cell tumor (PEComa), focusing on the analysis of its imaging features in CT and MRI to enhance understanding and awareness of this rare tumor.

This article reports a case of a 27-year-old married woman who was found to have a pelvic mass two years ago. After detailed examination at our hospital, imaging studies, including enhanced CT and MRI, revealed a cystic-solid lesion in the left adnexal area, with preoperative considerations of ovarian cystadenoma or uterine leiomyoma. She was referred to a specialized obstetrics and gynecology hospital for surgery, and the postoperative pathology was diagnosed as PEComa. She has been undergoing regular follow-up at our hospital post-surgery. One year after the operation, her laboratory tests showed no significant abnormalities, and imaging studies did not reveal any signs of metastasis.

Uterine broad ligament PEComa is a rare tumor, and accurate imaging features and classification criteria can aid in improving preoperative diagnosis. A deeper understanding of the clinical and imaging characteristics of this rare disease is significant for enhancing diagnostic accuracy and treatment outcomes.

Ultrasound is routinely used for thyroid nodule diagnosis, yet distinguishing benign from malignant TI-RADS category 4 nodules remains challenging. This study has integrated two-dimensional ultrasound, shear wave elastography (SWE), and contrast-enhanced ultrasound (CEUS) features via machine learning to improve diagnostic accuracy for these nodules.

A total of 117 TI-RADS 4 thyroid nodules from 108 patients were included and classified as benign or malignant based on pathological results. Two-dimensional ultrasound, CEUS, and SWE were compared. Predictive features were selected using LASSO regression. Feature importance was further validated using Random Forest, SVM, and XGBoost algorithms. A logistic regression model was constructed and visualized as a nomogram. Model performance was assessed using receiver operating characteristic (ROC) analysis, calibration curves, and decision curve analysis (DCA).

Malignant nodules exhibited significantly elevated serum FT3, FT4, FT3/FT4, TSH, and TI-RADS scores compared to benign lesions. Key imaging discriminators included unclear boundaries, aspect ratio ≥1, low internal echo, microcalcifications on ultrasound; enhancement degree, circumferential enhancement, and excretion on CEUS; and elevated SWE values (Emax, Emean, Esd, etc.) and altered CEUS quantitative parameters (PE, WiR, WoR, etc.) (all P< 0.05). A nomogram integrating four optimal predictors, including Emax, FT4, TI-RADS, and ∆PE, demonstrated robust predictive performance upon validation by ROC, calibration, and DCA curve analysis.

The nomogram incorporating Emax, FT4, TI-RADS, and ∆PE showed high predictive accuracy, particularly for papillary carcinoma in TI-RADS 4 nodules. Its applicability may, however, be constrained by the single-center retrospective design and limited pathological coverage.

The multimodal ultrasound-based machine learning model effectively predicted malignancy in TI-RADS category 4 thyroid nodules.

We developed and validated a novel CT-based radiomics nomogram aimed at improving the differentiation between checkpoint inhibitor-related pneumonitis (CIP) and COVID-19 pneumonia, addressing the persistent clinical uncertainty in pneumonia diagnosis.

A total of 97 patients were enrolled. CT image segmentation was performed, extracting 1,688 radiomics features. Feature selection was conducted using variance thresholding, the least absolute shrinkage and selection operator (LASSO) regression, and the Select K Best methods, resulting in the identification of 33 optimal features. Several classification models (K-Nearest Neighbors [KNN], Support Vector Machine [SVM], and Stochastic Gradient Descent [SGD]) were trained and validated using a 70:30 split and fivefold cross-validation. A radiomics nomogram was subsequently developed, incorporating the radiomics signature (Rad-score) alongside clinical factors. It was assessed based on area under the curve (AUC), sensitivity, specificity, and decision curve analysis (DCA).

The SVM classifier exhibited the highest performance, achieving an AUC of 0.988 in the training cohort and 0.945 in the validation cohort. The constructed radiomics nomogram demonstrated a markedly improved predictive accuracy compared to the clinical model alone (AUC: 0.853 vs. 0.810 in training; 0.932 vs. 0.924 in validation). Calibration curves indicated a strong alignment of the model with observed outcomes, while DCA confirmed a greater net benefit across various threshold probabilities.

A radiomics nomogram integrated with radiomics signatures, demographics, and CT findings facilitates CIP differentiation from COVID-19, improving diagnostic efficacy.

Radiomics acts as a potential modality to supplement conventional medical examinations.

Colorectal cancer (CRC) is a leading cause of cancer-related morbidity and mortality. Accurate differentiation of tumor grade is crucial for prognosis and treatment planning. This study aimed to evaluate the diagnostic value of dual-source CT dual-energy technology parameters in distinguishing CRC differentiation grades.

A retrospective analysis was conducted on 87 surgically and pathologically confirmed CRC patients (64 with medium-high differentiation and 23 with low differentiation) who underwent dual-source CT dual-energy enhancement scanning. Normalized iodine concentration (NIC), spectral curve slope (K), and dual-energy index (DEI) of the tumor center were measured in arterial and venous phases. Differences in these parameters between differentiation groups were compared, and ROC curve analysis was performed to assess diagnostic efficacy.

The low-differentiation group exhibited significantly higher NIC, K, and DEI values in both arterial and venous phases compared to the medium-high differentiation group (P < 0.01). In the arterial phase, NIC, K, and DEI yielded AUC values of 0.920, 0.770, and 0.903, respectively, with sensitivities of 95.7%, 65.2%, and 91.3%, and specificities of 82.8%, 75.0%, and 75.0%, respectively. In the venous phase, AUC values were 0.874, 0.837, and 0.886, with sensitivities of 91.3%, 82.6%, and 91.3%, and specificities of 68.75%, 75.0%, and 73.4%. NIC in the arterial phase showed statistically superior diagnostic performance compared to K values (P < 0.05).

Dual-energy CT parameters, particularly NIC in the arterial phase, demonstrate high diagnostic accuracy in differentiating CRC grades. These findings suggest that quantitative dual-energy CT metrics can serve as valuable non-invasive tools for tumor characterization, aiding in clinical decision-making. Study limitations include its retrospective design and relatively small sample size.

NIC, K, and DEI values in dual-energy CT scans are highly effective in distinguishing CRC differentiation grades, with arterial-phase NIC showing the highest diagnostic performance. These parameters may enhance preoperative assessment and personalized treatment strategies for CRC patients.

Prenatal intervention with fetoscopic endoluminal tracheal occlusion (FETO) using a balloon can stimulate lung growth and improve neonatal survival for moderate and severe congenital diaphragmatic hernia (CDH). Quantitative parameters measured on magnetic resonance imaging (MRI) can guide the treatment of CDH and evaluate changes after FETO treatment.

We reported on five cases of isolated left congenital diaphragmatic hernia (CDH) in fetuses who underwent FETO surgery. We conducted a comparison of the MRI images before and after FETO treatment and analyzed the correlation between the observed changes and the clinical outcomes of the neonates after delivery.

MRI can precisely provide the anatomical details of CDH and quantitatively analyze changes in fetal lung volume before and after FETO surgery.

Cytotoxic lesions of the corpus callosum (CLOCC) are a rare clinical-radiological syndrome with an unclear specific pathogenesis, and cases occurring consecutively in siblings are exceptionally uncommon. This study reports two pediatric siblings with CLOCC (one experiencing two episodes), highlighting the potential role of genetic susceptibility in its pathogenesis. The findings contribute to the limited literature on familial CLOCC and recurrent cases in children.

Two brothers (aged 9 and 12) presented with sudden-onset aphasia and unilateral limb weakness, preceded by rhinorrhea. Magnetic resonance imaging (MRI) revealed reversible lesions in the splenium of the corpus callosum and bilateral frontoparietal white matter, consistent with CLOCC. Both patients received immunomodulatory therapy (e.g., corticosteroids, intravenous immunoglobulin) and symptomatic treatment, achieving full neurological recovery within approximately one week. The elder sibling had a recurrence two years later (when the patient was 14 years old) with similar imaging findings. Laboratory tests ruled out common infections, and cerebrospinal fluid analysis was unremarkable.

This case underscores CLOCC as a heterogeneous condition with possible genetic predisposition, as evidenced by its occurrence in siblings and recurrence in one sibling. While prognosis is generally favorable, the observed sibling clustering and individual recurrence suggest the need for further research into underlying genetic or immunological mechanisms.

Early detection of gastric cancer remains challenging for many of the current imaging techniques. Recent advancements in reduced field-of-view (rFOV) diffusion-weighted imaging (DWI) have shown promise in improving the visualization of small anatomical structures. This study aimed to evaluate and compare the diagnostic performance of rFOV DWI with multi-detector computed tomography (MDCT) and conventional full field of view (fFOV) DWI for detecting early gastric cancer (EGC).

This retrospective study included 43 patients with pathologically confirmed EGC. All participants underwent pre-treatment imaging, including CT scans and MRI with a prototype rFOV DWI and conventional fFOV DWI at 3 Tesla. Quantitative (signal-to-noise ratio [SNR], contrast-to-noise ratio [CNR]) and qualitative (subjective image quality) assessments were performed. Diagnostic performance was evaluated using receiver operating characteristic (ROC) curves and area-under-the-curve (AUC) analysis.

rFOV DWI demonstrated significantly higher SNR and CNR compared with fFOV DWI (P < 0.05). Subjective image quality scores were also superior for rFOV DWI (P < 0.05). In lesion detection, rFOV DWI showed higher sensitivity (0.705) than CT (0.636) and fFOV DWI (0.523). ROC analysis revealed that rFOV DWI had a higher AUC (0.829, 95% CI [0.764, 0.882]) than fFOV DWI (0.734, 95% CI [0.661, 0.798], P = 0.02) and a modest improvement over CT (0.799, 95% CI [0.731, 0.856], P = 0.51).

The findings suggest that rFOV DWI provides superior image quality and diagnostic accuracy for EGC detection compared with conventional fFOV DWI. While it showed a trend toward better performance than CT, further studies with larger cohorts are needed to validate these results.

rFOV DWI offers improved image quality and diagnostic performance for early gastric cancer detection compared with fFOV DWI, with a potential advantage over CT. This technique may enhance early diagnosis and clinical decision-making in gastric cancer management.

This study aimed to investigate nerve fiber bundle damage associated with spinocerebellar degeneration, a dominant inherited neurological disorder, using magnetic resonance imaging (MRI) with diffusion tensor imaging (DTI).

Four cases of spinocerebellar degeneration and ten matched healthy subjects were retrospectively enrolled. DTI software was used for processing and analysis.

All patients had an abnormal spinocerebellar ataxia (SCA) type 3 gene mutation, with cerebellar and brainstem atrophy, a decreased signal in the pons and projection fibers. Significant interruption and destruction were revealed in the midline of the cerebellar peduncle, cerebellar arcuate fibers, and the spinothalamic and spinocerebellar tracts. Significant (p <0.05) decreases were detected in FA values in the cerebellar peduncle (0.51±0.04 vs. 0.68±0.02), cerebellar arcuate fibers (0.37±0.08 vs. 0.51±0.05), spinothalamic tract (0.42±0.03 vs. 0.49±0.05), and spinocerebellar tract (0.44±0.06 vs. 0.52±0.06) compared with healthy controls. Compared with healthy controls, significant (p <0.05) increases were detected in ADC values in the cerebellar peduncle (0.84±0.11 vs. 0.67±0.03), cerebellar arcuate fibers (0.87±0.12 vs. 0.66±0.05), spinothalamic tract (0.89±0.13 vs. 0.70±0.03) within the brainstem, and spinocerebellar tract (0.79±0.07 vs. 0.69±0.06).

The MRI DTI technique provides sufficient information for studying spinocerebellar degeneration and for conducting further research on its etiology and diagnosis. Some limitations were present, including the retrospective and single-center study design, a limited patient sample, and enrollment of only Chinese patients.

The MRI DTI technique can clearly demonstrate the degree of damage to nerve fiber bundles in the cerebellum and the adjacent relationship between the fiber bundles entering and exiting the cerebellum in patients with spinocerebellar degeneration.

Accurate segmentation of small and irregular pulmonary nodules remains a significant challenge in lung cancer diagnosis, particularly in complex imaging backgrounds. Traditional U-Net models often struggle to capture long-range dependencies and integrate multi-scale features, limiting their effectiveness in addressing these challenges. To overcome these limitations, this study proposes an enhanced U-Net hybrid model that integrates multiple attention mechanisms to enhance feature representation and improve the precision of segmentation outcomes.

The assessment of the proposed model was conducted using the LUNA16 dataset, which contains annotated CT scans of pulmonary nodules. Multiple attention mechanisms, including Spatial Attention (SA), Dilated Efficient Channel Attention (Dilated ECA), Convolutional Block Attention Module (CBAM), and Squeeze-and-Excitation (SE) Block, were integrated into a U-Net backbone. These modules were strategically combined to enhance both local and global feature representations. The model’s architecture and training procedures were designed to address the challenges of segmenting small and irregular pulmonary nodules.

The proposed model achieved a Dice similarity coefficient of 84.30%, significantly outperforming the baseline U-Net model. This result demonstrates improved accuracy in segmenting small and irregular pulmonary nodules.

The integration of multiple attention mechanisms significantly enhances the model’s ability to capture both local and global features, addressing key limitations of traditional U-Net architectures. SA preserves spatial features for small nodules, while Dilated ECA captures long-range dependencies. CBAM and SE further refine feature representations. Together, these modules improve segmentation performance in complex imaging backgrounds. A potential limitation is that performance may still be constrained in cases with extreme anatomical variability or low-contrast lesions, suggesting directions for future research.

The Enhanced U-Net hybrid model outperforms the traditional U-Net, effectively addressing challenges in segmenting small and irregular pulmonary nodules within complex imaging backgrounds.

Accurate liver volumetry is crucial for hepatectomy. In this study, we developed and validated a deep learning system for automated liver volumetry in patients undergoing hepatectomy, both preoperatively and at 7 days and 3 months postoperatively.

A 3D U-Net model was trained on CT images from three time points using a five-fold cross-validation approach. Model performance was assessed with standard metrics and comparatively evaluated across the time points.

The model achieved a mean Dice Similarity Coefficient (DSC) of 94.31% (preoperative: 94.91%; 7-day post-operative: 93.45%; 3-month post-operative: 94.57%) and a mean recall of 96.04%. The volumetric difference between predicted and actual volumes was 1.01 ± 0.06% preoperatively, compared to 1.04 ± 0.03% at other time points (p < 0.05).

This study demonstrates a novel capability to automatically track post-hepatectomy regeneration using AI, offering significant potential to enhance surgical planning and patient monitoring. A key limitation, however, was that the direct correlation with clinical outcomes was not assessed due to constraints of the current dataset. Therefore, future studies using larger, multi-center datasets are essential to validate the model's clinical and prognostic utility.

The developed artificial intelligence model successfully and accurately measured liver volumes across three critical post-hepatectomy time points. These findings support the use of this automated technology as a precise and reliable tool to assist in surgical decision-making and postoperative assessment, providing a strong foundation for enhancing patient care.

Heart failure (HF) is classified into three types based on left ventricular ejection fraction (LVEF). A newly developed transthoracic three-dimensional (3D) echocardiography Heart-Model (HM) offers quick analysis of the volume and function of the left atrium (LA) and left ventricle (LV). This study aimed to determine the value of the HM in HF patients.

A total of 117 patients with HF were divided into three groups according to EF: preserved EF (HFpEF, EF ≥50%), mid-range EF (HFmrEF, EF =41%–49%), and reduced EF (HFrEF, EF ≤40%). The HM was applied to analyze 3D cardiac functional parameters. LVEF was obtained using Simpson’s biplane method. The N-terminal pro-B-type natriuretic peptide (NT-proBNP) concentration was measured.

Significant differences in age, female proportion, body mass index, and comorbidities were observed among the three groups. With decreasing EF across the groups, the 3D volumetric parameters of the LA and LV increased, while LVEF decreased. The LV E/e' was significantly higher in HFrEF patients than in HFpEF patients. LVEF measurement was achieved in significantly less time with the HM compared with the conventional Simpson’s biplane method. The NT-proBNP concentration increased in the following pattern: HFrEF > HFmrEF > HFpEF. The NT-proBNP concentration correlated positively with LV volume and negatively with LVEF from both the HM and Simpson’s biplane method.

LA and LV volumes increase, and the derived LV systolic function decreases with increasing HF severity determined by the HM. The functional parameters measurements provided by the HM are associated with laboratory indicators, indicating the feasibility of using the HM in routine clinical application.

Deep learning methods were employed to perform harmonization analysis on whole-brain scans obtained from 1.5-T and 3.0-T scanners, aiming to increase comparability between different magnetic resonance imaging (MRI) scanners.

Thirty patients evaluated in Beijing Tsinghua Changgung Hospital between August 2020 and March 2023 were included in this retrospective study. Three MRI scanners were used to scan patients, and automated brain image segmentation was performed to obtain volumes of different brain regions. Differences in regional volumes across scanners were analyzed using repeated-measures analysis of variance. For regions showing significant differences, super-resolution deep learning was applied to enhance consistency, with subsequent comparison of results. For regions still exhibiting differences, the Intraclass Correlation Coefficient (ICC) was calculated and the consistency was evaluated using Cicchetti's criteria.

Average whole-brain volumes for different scanners among patients were 1152.36mm³ (SD = 95.34), 1136.92mm³ (SD = 108.21), and 1184.00mm³ (SD = 102.78), respectively. Analysis revealed significant variations in all 12 brain regions (p<0.05), indicating a lack of comparability among imaging results obtained from different magnetic field strengths. After deep learning-based consistency optimization, most brain regions showed no significant differences, except for six regions where differences remained significant. Among these, three regions demonstrated ICC values of 0.868 (95%CI 0.771-0.931), 0.776 (95%CI 0.634-0.877), and 0.893 (95%CI 0.790-0.947), indicating high reproducibility and comparability.

This study demonstrates a deep learning-based harmonization method that effectively mitigates field strength-related inconsistencies between 1.5-T and 3.0-T MRI, significantly enhancing their comparability. The high ICCs observed in key brain regions confirm the robustness of this approach, paving the way for reliable clinical application across different scanners. A noted limitation is its current focus on brain imaging, which warrants future research to extend its applicability to other anatomical areas.

This study employed a novel machine learning approach that significantly improved the comparability of imaging results from patients using different magnetic field strengths and various models of MRI scanners. Furthermore, it enhanced the consistency of central nervous system image segmentation.