Objective To evaluate the potential of deep learning image reconstruction (DLIR) in improving image quality and reducing radiation dose by comparing the noise power spectrum, task-based transfer function and lesion detection capability. Methods The ACR464 phantom was scanned using GE Revolution APEX CT and eight different noise indices (NI=10, 14, 16, 18, 20, 22, 24, 28) were set. The original data were subjected to image reconstruction using filtered back-projection (FBP), multi-model iterative reconstruction algorithms (ASiR-V) at 40%, ASiR-V at 60%, ASiR-V at 80%, and different levels of deep learning image reconstruction (DLIR-L, DLIR-M, DLIR-H) algorithms. The image quality was evaluated by using imQuest software to calculate the noise power spectrum (NPS), task-based transfer function (TTF), and detection capability index (d') of different reconstruction algorithms. Results Among all the reconstruction algorithms, the NPS peak of DLIR-H was the lowest. With the increase of noise index, both NPS and fav move towards low frequencies. The fav of DLIR-H (0.24-0.27 mm^-1) was only 40% lower than that of ASiR-V (0.26-0.28 mm^-1). The TTF_50% value was not affected by the DLIR level. The TTF_50% value was (37.44±10.85)% and (46.24±15.28)% higher than that of ASiR-V60% and 80%, respectively. The detection ability of both large and small features in deep learning image reconstruction was 40% higher than that of ASiR-V. When comparing the radiation doses with comparable lesions detection capabilities of 40% ASiR-V at NI=10 and DLIR-H, the radiation dose for small features decreased by approximately 76.48%, and that for large features decreased by approximately 72.59%. Conclusion Deep learning image reconstruction can not only reduce noise, improve spatial resolution and lesion detectibility without changing noise texture, but also has more powerful ability to reduce radiation dose than ASiR-V.

With the extension of life expectancy, population aging and the increasing healthcare demands of elderly patients have become pressing realities. Aged patients often experience diminished physiological reserve and progressive deterioration across multiple systems including physiological, metabolic, and sensory functions resulting in reduced tolerance to intravenous anesthetics. Ciprofol is a novel non-barbiturate intravenous anesthetic independently developed and recently approved in China. Existing evidence indicates that ciprofol exhibits characteristics such as rapid onset, faster recovery, reduced injection pain, higher potency, and a wider safety window. However, most studies to date are small, single-center investigations lacking rigorous evidence grading, particularly in the context of clinical application in elderly patients. Based on current research, this review summarizes the clinical use of ciprofol in the elderly population, with the aim of supporting its rational administration in this demographic and proposing directions for future research.

Vascular cognitive impairment (VCI) is a clinical syndrome characterized by cognitive impairment in at least one domain, caused by cerebrovascular lesions and their risk factors, with an increasing incidence among middle-aged and elderly populations. In recent years, advancements in MRI have provided new methods for the study of VCI. Among these, diffusion tensor imaging, which is currently the only non-invasive technique that allows visualization of the arrangement of white matter fiber tracts in living brains, has unique advantages in VCI research. This article will review the concept of diffusion tensor imaging and its various applications in VCI, including white matter lesions, diagnosis and classification, risk factor identification, and prevention and treatment, while also discussing its limitations and prospects for future development. This aims to provide new insights for the clinical diagnosis and pathological research of neurological diseases.

Objective To explore the value of a Nomogram model based on CT radiomics with clinical parameters in differentiating non-small cell lung cancer from benign pulmonary lesions. Methods A retrospective study was conducted on 177 patients with benign pulmonary lesions and non-small cell lung cancer, confirmed by pathology, at the First Affiliated Hospital of Bengbu Medical Unversity from December 2020 to December 2023. The cases were randomly divided into a training group and a validation group in an 8:2 ratio. Radiomic features were extracted from contrast-enhanced CT images, and a stepwise dimensionality reduction was performed using the Relief-LASSO algorithm, ultimately selecting five optimal features from a total of 2264 radiomic features. Single and multiple factor Logistic regression was employed to screen independent clinical risk factors. Clinical, radiomics, and Nomogram models were constructed respectively. The performance of the Nomogram model was comprehensively evaluated using multiple metrics, including the area under the ROC curve (AUC), calibration curves, and decision curve analysis. Results The results indicated that the Nomogram model exhibited excellent predictive performance, with AUC values of 0.872 (95% CI: 0.817-0.928) in the training set and 0.788 (95% CI: 0.627-0.948) in the validation set. These values were significantly higher than those of the individual imaging model (0.811, 0.722) and the clinical model (0.797, 0.734). Conclusion The established Nomogram model serves as a non-surgical predictive tool for the differential diagnosis of non-small cell lung cancer and benign pulmonary lesions. Validation demonstrated that the Nomogram model exhibited excellent differentiation and calibration abilities, indicating its clinical utility in the early screening of lung cancer and providing important guidance for clinical decision-making prior to surgery.

Objective To explore the clinical value of a combined model constructed based on ultrasound radiomics features and clinicopathological features in predicting pathological complete response after neoadjuvant chemotherapy (NAC) for breast cancer. Methods Retrospective analysis included 160 breast cancer patients receiving NAC before surgery at the First Affiliated Hospital of Bengbu Medical College from August 2021 to July 2024. By analyzing the ultrasound images, the radiomics features were extracted, and the features with statistical significance were screened out for subsequent modeling. Multivariate Logistic regression analysis and the XGBoost algorithm were used to construct joint models based on ultrasound radiomics features, clinicopathological features, and their combination, respectively, and nomograms were drawn. Model performance for NAC response was evaluated using AUC and decision curve analysis, with calibration curves assessing prediction-actuality agreement in the combined model. Results Estrogen receptor (P=0.006), progesterone receptor (P=0.024), and human epidermal growth factor receptor-2 (P=0.034) emerged as independent predictors of NAC response among clinicopathological features. The integrated model combining these three clinical predictors with radiomics features demonstrated optimal predictive performance, achieving a training-set AUC of 0.863, significantly higher than individual models. Calibration curves confirmed robust agreement between predicted and observed outcomes, while decision curve analysis indicated substantial clinical net benefit. Conclusion The combined model constructed based on the characteristics of ultrasound radiomics and clinicopathological features can effectively predict the efficacy of NAC in breast cancer and provide strong support for clinical decision-making. This model has high predictive accuracy and clinical application value, and is expected to become an important tool for individualized treatment of breast cancer.

Objective To construct a machine learning model using multimodal MRI radiomics features and clinical characteristics to predict lymph node metastasis in rectal cancer patients. Methods A retrospective analysis was conducted on clinical data and MRI images of 223 rectal cancer patients treated at Shunde Hospital of Southern Medical University from May 2018 to May 2023. Tumor regions were delineated using 3D Slicer software, and radiomics features were extracted using PyRadiomics software. Univariate logistic regression and LASSO regression were used to screen features, and clinical, radiomics, and clinico-radiomics models were constructed and evaluated in the training cohort (n=157) and validation cohort (n=66) to evaluate their predictive performance for lymph node metastasis. Results In the training cohort, the AUC values for the clinical, radiomics, and clinico-radiomics models were 0.668, 0.725, and 0.771, respectively. The radiomics model and clinico-radiomics model demonstrated good predictive performance in the validation cohort, with the radiomics model achieving an AUC of 0.780. Conclusion Machine learning models based on MRI radiomics and clinical features can effectively predict lymph node metastasis in rectal cancer. Radiomics features provide high predictive value, while clinical features offer limited predictive value.

Objective To investigate the application value of cardiac magnetic resonance imaging with late gadolinium enhancement (CMR-LGE) in the diagnosis of dilated cardiomyopathy (DCM), and to compare the diagnostic results of CMR-LGE with those of echocardiogram (UCG). Methods A retrospective study was conducted on 36 patients diagnosed with DCM upon discharge from Anhui Second Provincial Hospital from January 2023 to March 2025. All patients underwent both CMR-LGE (observation group) and UCG (control group) examinations. Within the observation group, 29 patients were CMR-LGE positive for DCM (observation subgroup 1), and 7 were CMR-LGE negative (observation subgroup 2). Within the control group, 20 patients were UCG positive for DCM (control subgroup 1), and 16 were UCG negative (control subgroup 2). Differences in diagnostic results, CMR-derived parameters, and clinical specificity parameters were analyzed between the groups. Results CMR-LGE confirmed the diagnosis in 29 cases, which was significantly higher than the 20 cases diagnosed by UCG (P=0.014). In observation subgroup 1 (CMR-LGE positive), the left ventricular end-diastolic volume index, left ventricular end-systolic volume index, and myocardial mass index were significantly higher than those in observation subgroup 2 (CMR-LGE negative). Conversely, body surface area, left ventricular ejection fraction, and stroke volume index were significantly lower in observation subgroup 1 compared to observation subgroup 2 (P0.05). ROC curve analysis demonstrated that SVI had the highest diagnostic efficacy for DCM, with an AUC of 0.990. Conclusion CMR-LGE demonstrates higher sensitivity in diagnosing DCM compared to UCG. Its quantitative parameters provide valuable reference for clinical diagnosis and treatment.

Lung cancer, characterized by its complex and diverse etiology and subtle early clinical symptoms, has led to a consistently high incidence rate and a continuous increase year after year.CT imaging technology, due to its extremely high detection rate and ability to non-invasively and conveniently identify suspected lung lesions, is currently the best method for early lung cancer screening. This review will systematically elaborate on the advantages and disadvantages of two-dimensional reconstruction technology and three-dimensional reconstruction technology in CT images, and focus on the analysis of the clinical application value of three-dimensional reconstruction technology in the diagnosis and treatment of pulmonary nodules in chest CT images.The purpose is to select the most suitable reconstruction method for the diagnosis of clinical lung diseases, and to provide a reference for the innovation of three-dimensional reconstruction technology of lung nodules in chest CT in the future.

The Ki-67 index in breast cancer reflects the proliferative activity of tumor cells. High Ki-67 expression indicates greater proliferative activity, increased invasiveness, higher recurrence risk, and poorer prognosis. Therefore, Ki-67 expression status is essential for molecular subtyping, assessment of treatment efficacy, and prognosis prediction in breast cancer. Radiomics and deep learning have become prominent approaches in intelligent medical imaging, enabling comprehensive, non-invasive, and dynamic assessment of Ki-67 expression in breast cancer by extracting high-throughput imaging features that reflect tumor heterogeneity. This review summarizes recent progress in applying radiomics and deep learning to predict Ki-67 expression status in breast cancer.

Objective To construct a dual-phase integrated enhanced CT radiomics model for predicting PD-L1 expression in non-small cell lung cancer (NSCLC) patients. Methods A retrospective study was conducted on 150 NSCLC patients who were pathologically confirmed at the Affiliated Hospital of Jining Medical University from November 2019 to July 2023. These patients were randomly assigned to a training cohort (105 cases) and a testing cohort (45 cases) at a ratio of 7:3. Radiomics features were extracted from both the arterial and venous phases of CT images. Dimensionality reduction and key feature selection were performed using the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm. Eight machine learning algorithms, including logistic regression, were employed to construct radiomics models. The best predictive model was identified through ROC curve analysis, and a dual-phase integrated radiomics model was developed by combining radiomics features from both phases. Univariate and multivariate logistic regression analyses were conducted to evaluate clinical features and to identify independent predictors for constructing a clinical model. A Combine model was then established by integrating radiomics and clinical features. The performance of the models was assessed using ROC curves, and their clinical utility was evaluated using decision curve analysis. Results A total of 1835 radiomics features were extracted from both the arterial and venous phase CT images. After dimensionality reduction and selection, 9 radiomics features were ultimately chosen from each phase. Among the radiomics models, the logistic regression model exhibited higher predictive efficiency and robustness. The dual-phase integrated enhanced CT radiomics model demonstrated superior performance compared to single-phase models. The radiomics-clinical model showed the best discriminative ability, with AUC values of 0.822 in the training cohort and 0.681 in the testing cohort. Decision curve analysis indicated the best clinical effectiveness. Conclusion The diagnostic model combining radiomics and clinical features of NSCLC has a good ability to predict PD-L1 expression and can provide a non-invasive and effective diagnostic method for clinical practice.

Objective To explore the predictive value of imaging features, serological immune indicators, and clinical features for interstitial lung disease in Sj?gren's syndrome (SS-ILD). Methods A total of 161 patients with Sj?gren's Syndrome (SS) who were treated at the First Affiliated Hospital of Henan Medical University from July 2018 to February 2025 were retrospectively enrolled; through propensity score matching (PSM), the patients were divided into the SS without interstitial lung disease (ILD) group (n=56) and the SS with ILD group (n=29). Univariate analysis was used to screen for meaningful variables, the Spearman correlation coefficient was applied to analyze the correlation between these variables and SS-ILD, binary logistic regression was performed to construct a combined prediction model, the area under the receiver operating characteristic curve (AUC) and Delong test were used to verify the predictive performance of the combined model, a calibration curve was plotted to validate the value of imaging indicators incorporated into the combined model. Results Statistically significant differences were observed between the two groups in mean spleen density, percentage of CD4?T cells, complement C4, Dry mouth, Dry eyes, and oral symptoms, and these variables were all negatively correlated with the occurrence of SS-ILD (all P<0.05); Logistic regression analysis revealed that the mean splenic density (AUC=0.698, sensitivity=0.759, specificity=0.607), percentage of CD4⁺ T lymphocytes (AUC=0.677, sensitivity=0.982, specificity=0.310), and oral symptoms (AUC=0.635, sensitivity=0.339, specificity=0.931) all had independent predictive value. The three-factor combined model (AUC=0.781, sensitivity 0.862, specificity 0.554) was significantly superior to any single indicator (P<0.05). Conclusion Decreased mean splenic density, reduced percentage of CD4? T lymphocytes, and absence of oral symptoms are independent predictive factors for the development of SS-ILD. The combination of these three factors can effectively improve the predictive efficacy with high sensitivity, which is helpful for the early clinical identification of high-risk patients with SS-ILD.

Objective To analyze the imaging features of pediatric diffuse midline glioma H3K27M variant (DMG-A). Methods The data of 22 children with DMG-A admitted to our hospital from February 2019 to October 2023 were retrospectively analyzed. Preoperative CT scan was performed in 12 cases (11 cases simultaneously with MRI). The images were independently read by two experienced pediatric imaging diagnostic doctors with over 10 years of working experience. Observed the lesion location, shape, size, density/signal characteristics of the solid part, signal intensity, restricted diffusion and MRS Manifestations, intratumoral calcification, hemorrhage and necrosis/cystic changes, enhancement features, whether there are blood vessels passing through the tumor, peritumoral edema, adjacent and secondary changes of the tumor, whether there is distant metastasis, and the metastatic site, etc. Results CT examinations were conducted in 10 cases. The brainstem lesions were mostly low/slightly low density shadows (9 cases), followed by equal or slightly high density shadows (1 case each). It is manifested as thickening of the brainstem, mostly centered on the pons, and partially involving the medulla oblongata, midbrain, cerebellar peduncle and cerebellar hemispheres. The thalamic lesion was a mass-like, uneven, slightly high-density shadow (1 case), extending towards the cisterns and the right side of the midbrain. MRI examination was conducted in 21 cases. The brainstem lesions were characterized by significant thickening and enlargement of the brainstem, with the pons being the most prominent. The main signal shadows were low on T1WI and slightly high on T2WI/FLAIR (15 cases). DWI could have diffuse limitation (9 cases), or no diffuse limitation (5 cases), and another case had no DWI examination. The lesions mostly surrounded the basilar artery (14 cases). Thalamic lesions were manifested as an increase in thalamic volume (6 cases), mainly low signal shadows on T1WI (5 cases), followed by isosignal shadows (1 case), and slightly high signal shadows on T2WI/FLAIR. DWI may have diffuse limitation (4 cases) or no limitation (2 cases). It partially affects the cerebellum, midbrain and other parts downward. After enhancement, most cases presented with obvious nodular, flower-shaped and patchy enhancement (13 cases), and could also show mild heterogeneous enhancement and no obvious enhancement (3 cases each). Additionally, 2 cases had no MRI enhancement examination. Nine cases underwent MRS Examination. The main manifestations were elevated Cho and Cr peaks, decreased NAA peak, and elevated Cho+Cr/NAA. Conclusion Although the imaging manifestations of DMG-A in children are varied, they still have certain characteristics. A comprehensive analysis of the age of occurrence, location of onset, whether the diffusion is limited, and the way and extent of intensification is helpful to improve the imaging diagnosis and differential diagnosis of the disease.

Objective To evaluate the clinical, pathological, and imaging characteristics of multiple renal hemangiomas in different renal functions, namely end-stage renal disease (ESRD) and non-ESRD. Methods A retrospective analysis was performed on 1 case of ESRD-related multiple renal hemangioma confirmed by surgery in Shanxi Bethune Hospital (Shanxi Academy of Medical Sciences) on April 2017. Additionally, 23 cases of pathologically confirmed multiple renal hemangiomas patients from 1980 to 2023 were collected from the PubMed database. They were divided into the ESRD group and non-ESRD group according to renal function status. The clinical basic conditions, imaging features, treatment methods, and prognosis of the two groups were analyzed. Results The ESRD group included 19 cases (79.2%) , and the non-ESRD group included 5 cases (20.8%). Males were more common. There were no statistically significant differences between the two groups in terms of age, gender, clinical symptoms, tumor side, location, and concomitant renal epithelial tumors (P0.05). Extramedullary hematopoiesis was only observed in the ESRD group, with a statistically significant difference (P0.05). Except for the 1 case in our hospital, 5 patients were followed up for an average of 16 months, and no signs of tumor recurrence or metastasis were found. Conclusion Multiple renal hemangiomas are rare and mostly occur in the kidneys with ESRD. There were no significant differences in gender, age, clinical features, and imaging manifestations between multiple renal hemangiomas in ESRD and non-ESRD, but there were differences in extramedullary hematopoiesis.

Breast cancer is one of the most common types of cancer in women. The status of axillary lymph nodes plays a key role in clinical staging, treatment planning and prognosis evaluation of malignant tumors. At present, sentinel lymph node biopsy and axillary lymph node dissection are the gold standards for axillary lymph node assessment. Although it is widely used in clinical practice, its traumatic operation may cause a variety of postoperative complications. Therefore, evaluating the status of axillary lymph nodes by non-invasive methods before the operation is of great significance for formulating clinical diagnosis and treatment plans. Ultrasound imaging technology can precisely and non-invasively assess the status of axillary lymph nodes in breast cancer without radiation, and it is the main method for preoperative clinical assessment of the status of axillary lymph nodes in breast cancer. This article reviews the research progress of two-dimensional ultrasound, color Doppler flow imaging, elastography, contrast-enhanced ultrasound, ultrasound radiomics and deep learning techniques in predicting the status of axillary lymph nodes before breast cancer surgery, with the aim of providing a basis for formulating precise individualized treatment plans in clinical practice.

Objective This research focused on the synthesis and characterization of a dual-modal nanoprobe (Gal-MnO₂/CDDP@PDA-Cy5.5) for fluorescence and magnetic resonance imaging of hepatocellular carcinoma. The study further explored the probe's specific targeting efficacy against ASGPR-expressing Huh-7 cells and evaluated its performance in magnetic resonance imaging through in vitro experiments. Methods Potassium permanganate solution was added to silica and etched with anhydrous sodium carbonate. The resulting solution was conjugated with polydopamine (DA), galactosamine (Gal), and fluorescent Cy5.5 to obtain Gal-MnO₂@PDA-Cy5.5. Cisplatin (CDDP) was introduced into the solution, followed by overnight incubation at room temperature and subsequent centrifugation to remove unreacted CDDP. Cytotoxicity and cellular uptake were assessed using the CCK-8 assay and flow cytometry, respectively. The relaxation rate was measured by Niumag small-scale NMR spectroscopy, and the enhancement degree was evaluated using magnetic resonance imaging. Results The prepared nanoprobes, as observed by transmission electron microscopy, exhibited uniform size and granular morphology with a particle size of 185.0±6.3 nm. The hydrodynamic diameter was measured to be 185.1±16.4 nm, with a zeta potential of 22.5±0.3 mV and a relaxivity of 14.589 (mmol/L)^-1s^-1. The magnetic resonance imaging signal intensity progressively enhanced with increasing nanoprobe concentration. Cytotoxicity assays demonstrated minimal toxicity of the nanoprobes. Conclusion This study successfully synthesized a dual-modal nanoprobe targeting ASGPR, designated as Gal-MnO₂/CDDP@PDA-Cy5.5, which demonstrates specific binding to target cells in vitro. Validation results confirmed that the synthesized nanoprobe exhibits excellent stability and biosafety, along with specific targeting capability toward Huh-7 cells. Furthermore, it possesses imaging functionality that enhances T1 contrast in magnetic resonance imaging.

Objective To develop a preoperative model combining MRI features and clinical variables for predicting high Ki-67 expression in hepatocellular carcinoma (HCC) and to assess its prognostic value. Methods A total of 344 patients with solitary HCC who underwent hepatectomy at Zhongshan Hospital, Fudan University from January to December 2020 were retrospectively enrolled. Among them, 191 patients showed high Ki-67 expression (>25%) and 153 patients had low Ki-67 expression (≤25%) based on postoperative pathological findings. Preoperative MRI features, clinical variables, and pathological data were collected, and each HCC lesion was assigned a LI-RADS. The relationship between MRI, clinical, and pathological features and high Ki-67 expression was compared. The model's performance was evaluated using ROC curves, and the recurrence-free survival (RFS) was compared using the Kaplan-Meier method. Results Multivariable logistic regression identified age and lower alpha-fetoprotein (AFP) as protective factors, whereas high Edmondson-Steiner grade, corona enhancement and LI-RADS were independent risk factors for high Ki-67 (P<0.05). A composite score incorporating these five variables yielded an AUC of 0.747 (sensitivity 68.1%, specificity 71.9%), outperforming any single predictor (P<0.05). Overall RFS did not differ between high and low Ki-67 groups (P>0.05). Among patients with high Ki-67, those with microvascular invasion had significantly shorter RFS than those without microvascular invasion (P<0.05); no such difference was observed in the low Ki-67 subgroup (P>0.05). Conclusion Preoperative MRI features (corona enhancement, LI-RADS) combined with clinical variables (age, AFP, Edmondson-Steiner grade) reliably predict high Ki-67 expression in HCC and provide imaging evidence for prognostic stratification.

Objective To analyze the clinical characteristics of ovarian tumors during pregnancy and explore the diagnostic value of MRI examination in patients with ovarian tumors during pregnancy. Methods This retrospective study analyzed MRI findings and clinical data from 67 pregnant patients diagnosed with ovarian tumors during pregnancy confirmed by postoperative histopathology. Results A total of 82 ovarian tumors were identified in 67 patients, with 15 cases exhibiting bilateral involvement. Tumor diameters ranged from 2.0-34.7(9.3±4.9) cm. Histopathological subtypes included: 11 mature cystic teratomas, 11 mucinous cystadenomas, 10 serous cystadenomas, 9 endometriotic cysts, 7 luteinized cysts, 6 corpus luteum cysts, 3 simple cysts, 4 borderline tumors, 3 high-grade serous carcinomas, 1 sex cord-stromal tumor, 1 dysgerminoma, and 1 metastatic tumor. MRI demonstrated a diagnostic accuracy of 91.5%, with 7 cases misdiagnosed. Most ovarian tumors appeared as cystic or cystic-solid masses. Benign tumors predominantly presented as cystic lesions, whereas borderline and malignant tumors tended to show solid components. No statistically significant differences were observed between benign, borderline, and malignant tumor groups regarding tumor size, presence of pedicle torsion, or elevation of tumor markers (P0.05). In terms of clinical management and outcomes: 7 patients underwent pregnancy termination during the first trimester due to absence of fetal heart tones or lower abdominal pain; 22 patients with benign tumors underwent surgery during the second trimester, primarily via laparoscopy, due to abdominal pain or large tumor size; 30 patients with benign or borderline tumors were monitored via ultrasound and underwent tumor resection during term cesarean section. Among 6 patients with malignant tumors, 1 terminated the pregnancy due to fetal chromosomal abnormalities, 2 experienced preterm delivery following fetal lung maturation and subsequently underwent radical surgery, and 3 underwent fertility-sparing surgery during the second trimester. Overall, maternal and fetal outcomes were favorable. Conclusion Ovarian tumors during pregnancy are predominantly benign and asymptomatic. MRI is a valuable tool for the qualitative assessment of these tumors, supporting accurate diagnosis and informed clinical decision-making.

Breast cancer is the malignant tumor with the highest incidence in women worldwide. Therefore, early detection and early diagnosis are the keys to prolong survival, but there are limitations in traditional imaging diagnostic methods. In recent years, artificial intelligence (AI) technology has significantly improved the accuracy and efficiency of breast cancer imaging diagnosis through deep learning and image processing. In breast ultrasound, mammography, Breast MRI and emerging imaging technologies, AI can promote the development of medical imaging through lesion detection, classification, image enhancement, risk prediction and clinical decision support. However, the clinical translation of AI still faces challenges such as data standardization, algorithm generalization, and ethical compliance. In the future, it is necessary to strengthen multi-center cooperation, promote technological innovation, and improve ethical regulations, to ensure that it can truly meet clinical needs, so as to promote the intelligence, precision and universality of breast cancer diagnosis and treatment. This paper provides a systematic review of the comparative advantages and limitations of AI versus conventional imaging methods in the diagnosis, treatment, and prognosis prediction of breast diseases. It explores feasible pathways for clinical translation and future development directions, while also offering insights into the prospective applications of AI in breast disease diagnosis and treatment.

With the acceleration of global population aging, sarcopenia has developed into a significant public health problem that seriously affects the health and quality of life of the elderly. The refined application of MRI provides a new perspective to reveal the microscopic pathological changes in the muscle. Advanced MRI techniques, including Dixon imaging, DTI, T2-mapping, and MRS, enable the quantitative assessment of critical parameters such as intramuscular fat infiltration, fibrosis, and energy metabolism. These methodologies provide a robust theoretical foundation for the early diagnosis, mechanistic analysis, and intervention strategies for sarcopenia. Moreover, sleep and physical activity serve as fundamental pillars for skeletal muscle recovery, repair and growth, promoting anabolic hormone secretion, optimizing protein synthesis conditions, inhibiting catabolic processes, and reducing inflammation. Consequently, by elucidating the intrinsic relationship between sleep, physical activity, and muscle health, and integrating MRI as a non-invasive and quantitative assessment tool for muscle mass, composition and structural changes, this paper establishes a rigorous scientific basis for the early detection, risk stratification, and personalized management of sarcopenia.

Breast cancer remains one of the leading causes of cancer-related mortality among women worldwide. Neoadjuvant chemotherapy (NAC) has become an essential component of contemporary breast cancer treatment. However, conventional methods for assessing NAC efficacy are often limited by subjectivity and suboptimal accuracy, underscoring the urgent need for more objective and reliable evaluation strategies. In recent years, AI, particularly radiomics and deep learning, has driven significant advances in medical imaging analysis. MRI-based radiomics combined with DL has demonstrated the ability to extract high-dimensional features from imaging data that are invisible to the human eye. These features can capture subtle microstructural alterations within tumors, characterize biological phenotypes, and quantify intratumoral heterogeneity, thereby substantially improving the precision of treatment response evaluation. This review highlights recent progress in the application of MRI-based AI technologies for predicting NAC response in breast cancer, aiming to facilitate the translation of these techniques from theory to clinical practice and to provide a scientific foundation for advancing precision and personalized oncology care.

Objective To compare the differences in image quality between adaptive statistical iterative reconstruction (ASiR) and model-based iterative reconstruction (MBIR) techniques in routine-dose abdominal thin-slice CT scanning. Methods Twenty patients admitted to the Affiliated Hospital of Shaanxi University of Traditional Chinese Medicine from June to September 2024 who underwent CT scanning of the upper abdomen were selected to undergo conventional dose CT scanning of the abdomen using a GE Discovery CT 750HD, and the images were reconstructed with filtered back-projection reconstruction (FBP) and iterative reconstruction techniques after data acquisition,respectively, and the thickness of the reconstructed layer was set at 0.625 mm.The images were analysed and evaluated for parallel quality with the help of a post-processing platform. The images were analysed with the help of a post-processing platform for parallel quality evaluation,focusing on measuring and comparing the noise value (SD), contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) of the liver, spleen and muscle in each reconstruction mode. The reconstructed images were subjectively judged by two senior radiologists in a double-blind manner. Results Focusing on the liver,the liver noise of FBP, 40% ASiR and VEO3.0-NR40 reconstructed images were 23.45±3.34, 16.93±2.53, 5.72±1.02, respectively, and the noise of the 40% ASiR and VEO3.0-NR40 reconstructed images were reduced by 27.8% and 75.6%, respectively, compared with that of FBP (P0.001); CNR of the liver in the 3 groups of images was 9.25±1.86, 12.54±2.52, 32.83±4.61, and SNR was 3.82±0.77, 5.13±0.95, 13.48±2.34, respectively. The CNR of the reconstructed images with 40% ASiR and VEO3.0-NR40 increased by 35.6% and 255.0% respectively compared with the FBP group, and the SNR increased by 34.3% and 252.9% respectively (P0.001). The subjective scores of the images in the 3 groups were 2.90±0.31, 3.77±0.43, 4.83±0.38, respectively. Conclusion At equivalent doses, 40% ASiR and VEO 3.0-NR40 effectively reduce noise in abdominal CT images, thereby enhancing image quality,outperforming the conventional FBP reconstruction algorithm. Among these, the VEO3.0-NR40 reconstruction algorithm demonstrated exceptional performance, significantly optimizing image details and reducing noise interference.

Knee Osteoarthritis (KOA) is a common chronic joint disease that not only causes joint structural damage but also leads to significant changes in muscle morphology and function, such as muscle atrophy, fat infiltration, and decreased muscle strength, through the mechanical imbalance between lower limb muscles and joints. The mechanical imbalance between lower limb muscles and joints is regarded as one of the important factors contributing to KOA. In recent years, the rapid development of MRI technology has provided new tools for the comprehensive assessment of the morphology, structure, and function of lower limb muscles in patients with KOA. This review systematically elaborates on the technical principles of T1-weighted imaging, T2-weighted imaging, T2 mapping imaging, Dixon water-fat separation technology, MRI radiomics, and related cutting-edge technologies, and introduces their applications in the evaluation of lower limb muscle atrophy, fat infiltration, and decreased muscle strength. The purpose of this review is to systematically summarize the application progress of multimodal MRI technology in the evaluation of lower limb muscles in KOA patients, provide a scientific basis for the optimization of clinical diagnosis and treatment plans, and the formulation of rehabilitation strategies, and contribute to the improvement of precision medical care for KOA patients.

Deep vein thrombosis (DVT) is a condition caused by abnormal blood clotting within deep veins. As a common peripheral vascular disease, its annual incidence rate is approximately 0.1%. DVT fragments can easily lead to pulmonary embolism. Additionally, approximately 20%-50% of patients develop post-thrombotic syndrome later on, severely impacting their quality of life. Therefore, early recognition and management of DVT are crucial for preventing complications such as pulmonary embolism. This review summarizes various imaging diagnostic techniques for DVT, including DSA, US, CT, and MRI, alongside deep learning-based medical image analysis methods for thrombus detection. The aim is to provide evidence-based guidance for DVT prevention and treatment.

Objective To investigate the application value of magnetic resonance diffusion weighted imaging (MRI-DWI) combined with dynamic contrast enhanced MRI (DCE-MRI) in the diagnosis of prostate cancer (PCa). Methods A total of 106 patients with highly suspected PCa after clinical and laboratory examination were selected from the Department of Urology, Maanshan People's Hospital from January 2020 to December 2023. The patients were divided into PCa group (n=38) and benign group (n=68) according to the results of pathological examination. All patients underwent conventional MRI, DWI and DCE-MRI examinations, and multi-parameter image data were collected. The apparent diffusion coefficient (ADC), volume transfer constant (K^trans), interstitium-to-plasma rate constant (K_ep) and extravascular extracellular space volume fraction (V_e) were compared between the two groups. Multivariate logistic regression and ROC curve were used to analyze the diagnostic efficacy of each parameter for PCa. Results On the DWI sequence, most of the lesions in the PCa group showed high signal, and the ADC value was lower than that in the benign group (P<0.05). In DCE-MRI, the levels of K^trans and K_ep in PCa group were higher than those in benign group (P<0.05). There was no significant difference in V_e level between the two groups (P>0.05). Multivariate Logistic regression analysis showed that ADC value (OR=0.641), K^trans (OR=1.948) and K_ep (OR=1.790) were independent predictors of PCa (P<0.05). ROC curve analysis showed that the AUC of combined diagnosis of PCa was 0.944, compared with single detection (ADC: AUC=0.876; K^trans: AUC=0.802; K_ep: AUC=0.749), the difference was statistically significant (P<0.05). Conclusion MRI-DWI combined with DCE has high application value in the diagnosis of PCa, which can provide effective imaging support and reliable quantitative parameters for the early detection of lesions.

Objective To explore the value of a deep learning-based ultrasound radiomics nomogram in predicting Ki-67 expression levels in invasive breast cancer. Methods A retrospective single-center study was conducted, collecting complete preoperative clinical data and ultrasound images from 465 patients with pathologically confirmed invasive breast cancer at the First Affiliated Hospital of Bengbu Medical University from January to December 2024. Image acquisition was performed using Mindray Resona 7 and Samsung HS60 color Doppler ultrasound systems. Based on immunohistochemical results, patients were divided into high and low Ki-67 expression groups and randomly assigned to training (n=326) and validation (n=139) cohorts at a 7:3 ratio. ITK-SNAP software was used to segment tumors from the largest 2D ultrasound cross-sectional images, with interobserver consistency of ROI delineation assessed by ICC. Pyradiomics was employed to extract radiomics features from tumor tissues, and four deep learning networks were pretrained to construct clinical, ultrasound radiomics, fusion, and combined nomogram models. Diagnostic performance and clinical utility were evaluated using ROC curves, calibration curves, and decision curve analysis. Results Nineteen optimal ultrasound radiomics features and the DenseNet121 deep learning model showed the best performance (P<0.05). In the training cohort, the AUCs for the clinical model, ultrasound radiomics model, deep learning model, fusion model, and nomogram were 0.79 (95% CI: 0.74-0.84), 0.85 (95% CI: 0.81-0.90), 0.87(95% CI: 0.83-0.91), 0.94(95% CI: 0.91-0.97), and 0.95(95% CI: 0.93-0.98), respectively. In the validation cohort, the corresponding AUCs were 0.76(95% CI: 0.68-0.84), 0.78 (95% CI: 0.70-0.85), 0.81(95% CI: 0.74-0.88), 0.91(95% CI: 0.86-0.96), and 0.93(95% CI: 0.89-0.98). Conclusion The deep learning-based ultrasound radiomics nomogram can effectively predict Ki-67 expression in invasive breast cancer.

Objective To explore the feasibility of applying dual-flow injection technique combined with 70 kV tube voltage scanning in head and neck computed tomography angiography (CTA). Methods A total of 160 patients who underwent head and neck CTA examinations at the Medical Imaging Center of Taihe Hospital in Shiyan City were prospectively collected from July to November 2024 and randomly categorized into three groups (A, B, C): group A (n=60) underwent 70 kV tube voltage scanning with dual-flow injection; group B (n=40) underwent 100 kV tube voltage scanning and dual-flow injection; group C (n=60) underwent 100 kV tube voltage scanning with conventional injection. For each group, the CT values, background noise (SD), contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) of the aortic arch, common carotid artery, internal carotid artery, and D2 segment of middle cerebral artery were recorded. Additionally, the computed tomography dose index (CTDI) and dose length product (DLP) values of each patient were recorded; the subjective evaluation was performed by 2 senior physicians using double-blind scoring of all VR images. All data were statistically analyzed. Results group A showed significantly higher CT values in the aortic arch compared to group C (P0.05), while no significant differences were observed in the remaining vessel segments between group A and C (P0.05). In group A, the CT values of the aortic arch, bilateral common carotid arteries, bilateral internal carotid arteries, and bilateral middle cerebral arteries were all above 300 HU, meeting the diagnostic requirements. At the level of the bifurcation of the aortic arch, right common carotid artery, and the beginning of the right middle cerebral artery, the SNR and CNR values of group A were significantly lower than those of group C (P0.05), with no significant differences compared to group B (P0.05). The CTDI, DLP, and Effective dose(ED) values of group A were significantly lower than those of groups B and C, and the differences were statistically significant (P0.05). Specifically, compared to group C, group A showed a 77.41% reduction in CTDI, a 77.37% reduction in DLP, and a 77.78% reduction in ED. Additionally, the iodine contrast dose in group A was reduced by 40% compared to group C. Conclusion In head and neck CTA, the dual-flow injection technique combined with 70 kV low-tube-voltage scanning method can significantly reduce the radiation dose and pure iodine contrast agent usage without affecting the image quality and meeting the diagnostic requirements. The method is feasible in clinical examination.

Objective To evaluate the relationship between ultrasound-derived fat fraction (UDFF) and non-contrast multislice helieal computed tomography (MSCT) CT values, aiming to estimate hepatic fat content using UDFF. Methods A total of 140 patients from Beijing Tsinghua Changgung Hospital were prospectively enrolled between February 2024 AND June 2024. Based on the order of admission, the first 110 patients were assigned to the training set and the subsequent 30 patients to the validation set. All patients underwent both ultrasonography and upper abdominal MSCT within 7 days. The training set was categorized into a normal group (hepatic fat content<5.0%, n=68), a mild fatty liver group (hepatic fat content≥5.0% and <10.0%, n=30), and a moderate-to-severe fatty liver group (hepatic fat content≥10.0%, n=12). UDFF and CT values were measured in the right hepatic lobe using regions of interest (ROI) analysis. Correlation between UDFF and CT values was assessed, and comparisons were made among different fatty liver groups. Diagnostic performance of UDFF was evaluated using ROC curve analysis. Results In the training set, UDFF showed a strong negative correlation with CT values (r=-0.738, P<0.001), with statistically significant differences among groups (P<0.001). The UDFF values in the moderate-to-severe fatty liver group were significantly higher than those in the mild fatty liver and normal groups and normal groups (P<0.001). For diagnosing mild fatty liver, the area under the curve (AUC) was 0.661 (P=0.011、95% CI: 0.536~0.785), with a sensitivity of 43.3%, specificity of 85.3%, and a cutoff value of 6.1%. For moderate-to-severe fatty liver, the AUC was 0.921 (P<0.001、95% CI: 0.809~1.032), with a sensitivity of 91.7%, specificity of 93.3%, and a cutoff value of 10.4%. In the validation set, UDFF demonstrated an accuracy of 71.4%, sensitivity of 57.1%, specificity of 76.2% for mild fatty liver diagnosis, and an accuracy of 88.9%, sensitivity of 100%, specificity of 85.7% for moderate-to-severe fatty liver diagnosis. Conclusion UDFF exhibits a strong negative correlation with CT values in the quantitative diagnosis of fatty liver, and demonstrates high accuracy in diagnosing moderate-to-severe fatty liver.

Non-small cell lung cancer (NSCLC) is one of the leading causes of cancer-related deaths globally. Occult lymph node metastasis poses a significant challenge to the diagnosis, treatment planning, and prognosis of NSCLC patients. Early detection of occult lymph node metastasis through imaging can help select the optimal treatment method, thereby improving patient outcomes and ultimately contributing to reducing NSCLC-related mortality. This review discusses recent advancements in imaging features and radiomics for the early identification of occult lymph node metastasis in NSCLC, as well as the interpretability methods of radiomics.

Cervical cancer is a common malignant tumor in the female reproductive system, closely related to persistent infection with high-risk human papillomavirus. Radiotherapy is one of the important treatment methods for cervical cancer. However, while radiotherapy produces therapeutic effects, it may also cause damage to the pelvic bone marrow, leading to the occurrence of hematological toxicity. Accurate understanding and early prediction of the occurrence and grading of hematological toxicity are of great significance for optimizing treatment plans and improving prognosis. This article reviews the research progress on the mechanism, influencing factors, and imaging assessment and monitoring of hematological toxicity in radiotherapy for cervical cancer, aiming to provide references for future clinical practice.

Objective To explore the predictive value of head CT combined with platelet-to-lymphocyte ratio (PLR) and D-dimer (D-D) for rehabilitation effects of Brunnstrom stage-based functional training in stroke patients during convalescence. Methods A total of 126 stroke patients in convalescence who received Brunnstrom stage-based functional training in Zhejiang Rehabilitation Hospital from January 2023 to December 2024 were selected. After one month of treatment, the Fugl-Meyer Assessment (FMA) was used to evaluate rehabilitation effects, and patients were divided into the good effect group and the poor effect group. Head CT findings, PLR, and D-D level before rehabilitation training were compared between the two groups. The ROC curve was used to evaluate the predictive value of head CT combined with PLR and D-D for rehabilitation effects of Brunnstrom stage-based functional training in stroke patients during convalescence. Results Among the 126 stroke patients in convalescence, 79 (62.70%) showed good rehabilitation effects, with a FMA score of 49.91±1.02. 47 (37.30%) showed poor rehabilitation effects, with a FMA score of 43.57±1.13. The proportion of abnormal head CT findings, PLR, and D-D level in the poor effect group were higher than those in the good effect group (P0.05). Multivariate logistic regression analysis revealed that head CT findings, PLR, and D-D were independent factors influencing rehabilitation effects of Brunnstrom stage-based functional training in stroke patients during convalescence (P0.05). ROC curve analysis showed that the optimal cutoff values of PLR and D-D for predicting rehabilitation effects of Brunnstrom stage-based functional training in stroke patients during convalescence were 150.09 and 208.85 ng/mL, respectively. The AUC and specificity of combination of head CT, PLR and D-D for predicting rehabilitation effects of Brunnstrom stage-based functional training in stroke patients during convalescence were 0.822 and 89.87%, which were higher than those of any single indicator (P0.05). Conclusion Head CT findings, PLR, and D-D level are independent factors influencing rehabilitation effects of Brunnstrom stage-based functional training in stroke patients during convalescence. In addition, combined use of head CT, PLR and D-D can enhance the prediction of rehabilitation effects of Brunnstrom stage-based functional training in stroke patients during convalescence.

Objective To compare the diagnostic efficacy of ultrasound strain elastography and radiomics in predicting pathological response after neoadjuvant chemotherapy for breast cancer, and to explore the feasibility of their combined application. Methods A retrospective analysis was conducted on 100 breast cancer patients who received neoadjuvant chemotherapy at Zhongshan Hospital of Traditional Chinese Medicine from January 2021 to June 2024. Ultrasound images of breast cancer were collected before neoadjuvant chemotherapy, and the stiffness parameters and radiomics scores of the lesions were obtained using strain elastography and radiomics methods. After neoadjuvant chemotherapy, surgical resection was performed, and the patients were confirmed as pathological complete response (pCR) or non-pathological complete response (npCR). The diagnostic efficacy of individual and combined parameters was evaluated by plotting ROC curves, including the area under the curve (AUC), sensitivity, and specificity. Results Postoperative pathology was divided into pCR (n=41) and npCR (n=59). The lesion size, proportion of hard lesions, and strain rate FLR (fat/lesion ratio) in the pCR group were all lower than those in the npCR group (P<0.05). The AUC of FLR for predicting pCR was 0.833 (optimal cut-off value 14.3, sensitivity 78.05%, specificity 77.97%); the AUC of radiomics score was 0.825 (optimal cut-off value 1.92, sensitivity 82.93%, specificity 71.19%), and there was no difference in efficacy between the two (P=0.88). The AUC of the combined model increased to 0.914, which was significantly better than that of individual indicators (all P<0.01). Conclusion Strain elastography and radiomics have comparable value in the assessment of pCR, and their combined application can significantly improve diagnostic efficacy, providing a new strategy for non-invasive assessment of the efficacy of neoadjuvant chemotherapy.

Objective To investigate the differential diagnostic value of energy spectrum CT imaging combined with tumor markers on lung adenocarcinoma and squamous cell carcinoma. Methods The medical records of patients who visit to Guangyuan Central Hospital from May 2022 to May 2024 with suspected lung space-occupying lesions who needed to be confirmed by enhanced CT scan were retrospectively analyzed. The patients were pathologically confirmed as lung adenocarcinoma (n=78) and lung squamous cell carcinoma (n=30). energy spectrum CT imaging examination (GE Discovery 750 gemstone energy spectrum 64-slice CT scanner) and serum tumor marker detection were conducted on all patients. The energy spectrum CT imaging was compared between groups of patients. The levels of serum tumor markers were measured, and the diagnostic value of energy spectrum CT imaging combined with serum tumor markers on lung adenocarcinoma and squamous cell carcinoma was analyzed by ROC curve. Delong test was adopted to compare the difference in the area under the curve (AUC) between each single index and combined model. Results Compared with lung squamous cell carcinoma group, the proportions of peripheral type distribution, ground glass nodule and pleural depression in lung adenocarcinoma group were higher (P<0.05) while the proportions of vascular convergence sign and mediastinum invasion were lower (P<0.05). Serum tumor markers levels in lung adenocarcinoma group were lower than those in lung squamous cell carcinoma group (P<0.05). The AUC of combined detection of enhanced arterial phase CT value, venous phase CT value, carcinoembryonic antigen, neuron-specific enolase, cytokeratin 19 soluble fragment 21-1 and squamous cell carcinoma antigen was higher than that of single detection, and the above AUCs were 0.92 (0.854-0.965), 0.87 (0.787-0.924), 0.70 (0.834-0.953), 0.91 (0.834-0.953), 0.93 (0.859-0.968), 0.93 (0.859-0.968), 0.76 (0.668-0.837) and 0.98 (0.966-1.000) respectively, and the sensitivity and specificity of combined detection (0.97, 0.96) were higher than those of single detection (0.80, 0.80, 0.96, 0.90, 0.90, 0.96; 0.90, 0.83, 0.77, 0.87, 0.87, 0.57). Conclusion Energy spectrum CT imaging combined with tumor markers has differential diagnostic significance in the diagnosis of lung adenocarcinoma and lung squamous cell carcinoma.

Objective To systematically analyze the research dynamics and core progress of electroencephalography (EEG) applied to acupuncture in the past 20 years by using bibliometric methods, to reveal the development trend of the discipline, the international cooperation mode, and the potential of clinical translation. Methods Based on the Web of Science core ensemble database, 181 documents were included and visualized using CiteSpace software, including annual publication volume, national/regional collaborative networks, journal co-citation, keyword co-occurrence and clustering analyses, and combined with emergent words to detect the stage-by-stage evolution of research hotspots. Results The number of articles published each year has shown a steady upward trend, with China (excluding Taiwan Province of China) publishing the most articles (n=97), and the highest collaborative centrality in the UK (0.59). The high-frequency cited journals were concentrated in the fields of neuroimaging and alternative medicine. Keyword analysis showed that research hotspots gradually shifted from pain and electroacupuncture to brain network regulation, randomized controlled trials and multimodal imaging techniques, and clustering themes focused on acupuncture neuromodulation mechanisms and autonomic nervous system regulation. The emergent words indicate that brain modeling and functional connectivity have become emerging directions, and EEG technology confirms that acupuncture enhances alpha rhythm synchronization and improves brain network topology, providing electrophysiological evidence for clinical efficacy such as stroke rehabilitation. Conclusion The cross-study of EEG and acupuncture has shifted from single-mechanism exploration to complex system analysis, and in the future, it is necessary to deeply integrate the multimodal technologies such as EEG-fMRI, artificial intelligence algorithms, and clinical data, to construct a dynamic assessment system of acupuncture efficacy, and to promote the optimization of personalized treatment plans and the translation of precision medicine.

Objective To explore the application value of multi-slice spiral CT (MSCT) in the diagnosis and risk classification of gastrointestinal stromal tumor (GIST), and analyze image features of GISTs at different risk levels. Methods From January 2019 to December 2024, 104 patients with GIST admitted to China Rongtong Medical and Health Group Co., Ltd. Anqing 116 Hospital were selected. All of the patients underwent MSCT plain scan and multi-phase enhanced scan. With surgical and pathological results as the gold standard, cases at extremely low risk and low risk were included in the low-risk group, while those at intermediate risk and high risk were included in the high-risk group. The differences in MSCT features between the two groups, and the correlation between MSCT features and risk level were analyzed. The ROC curves were used to evaluate the predictive efficacy of MSCT features for high-risk GIST, and the predictive model was clinically validated. Results Multivariate logistic regression analysis identified tumor diameter, lesion morphology, growth pattern, and ulceration as independent influencing factors for risk stratification (P<0.05).A logistic regression model was constructed based on MSCT features, and ROC curve analysis results showed that the area under the curve of the model for evaluating the risk level of GIST was 0.951 (P<0.001). The sensitivity and specificity were 94.74% and 84.85%, indicating higher diagnostic efficacy. TheHosmer Lemeshow goodness of fit test was used to evaluate the calibration of the risk grading diagnostic prediction model for GIST patients, and the results showed that the fitting level of the GIST patient risk grading diagnostic prediction model was good (P=0.675, adjusted R²=0.754). Clinical verification shows that the sensitivity of this prediction model was 100%, the specificity was 88.89%, and the accuracy was 92.31%. The consistency between the prediction model and the actual clinical situation was relatively high (Kappa=0.831). Conclusion MSCT can effectively help evaluate the risk level of GIST based on features such as tumor diameter, shape, growth pattern, and ulcer. Among them, tumor diameter >5 cm, irregular shape, transmural growth, and ulcer are independent imaging indicators for predicting high-risk GIST, which provides an important basis for preoperative risk assessment in clinical practice.