Surgical treatment remains the cornerstone of treatment for advanced ovarian cancer patients, where achieving optimal cytoreductive surgery directly determines clinical prognosis. Imaging modalities serve as critical tools for preoperative evaluation of resectability, providing essential guidance for optimal surgical planning. The primary objective of preoperative imaging is to assess both the extent of tumor metastasis and the likelihood of complete resection, thereby maximizing surgical benefits and improving patient outcomes. Approximately 70% of ovarian cancer patients present with advanced-stage disease at initial diagnosis, often accompanied by extensive pelvic-abdominal metastases and peritoneal seeding, which significantly increases surgical complexity. Consequently, precise radiological assessment of tumor burden and critical organ involvement is imperative. Currently, computed tomography (CT) stands as the first-line imaging modality for preoperative evaluation due to its widespread availability and comprehensive visualization of peritoneal metastases and lymphadenopathy. However, CT demonstrates limited resolution for detecting subcentimeter peritoneal implants (<5 mm) and mesenteric infiltration. Magnetic resonance imaging (MRI), with its superior soft tissue contrast and multiparametric/multidirectional imaging capabilities, offers enhanced accuracy in evaluating pelvic organ invasion, particularly at critical sites like the rectosigmoid junction and bladder. Diffusion-weighted imaging (DWI) further improves detection rates for occult metastases and surgically challenging lesions. While ¹⁸F-FDG positron emission tomography (PET)/CT exhibits higher specificity for distant and lymph node metastases, its sensitivity for peritoneal dissemination does not surpass CT or MRI. Emerging advances in radiomics and artificial intelligence promise revolutionary progress in resectability assessment for advanced ovarian cancer. These technologies enable not only the extraction of quantitative imaging features reflecting tumor heterogeneity (combined with clinical data to construct predictive models) but also leverage deep learning algorithms for automated metastasis segmentation/identification and three-dimensional visualization modeling. Nevertheless, comprehensive surgical resectability evaluation must ultimately integrate patient fitness, surgeon expertise, and multidisciplinary team discussions to optimize decision-making.

Endometriosis-associated gynecological tumors represent a distinct disease category with unique imaging features and management considerations. These tumors are categorized into two groups: endometriosis-associated ovarian cancer (EAOC) and extraovarian endometriosis-associated malignancies. The most sensitive magnetic resonance imaging (MRI) feature for diagnosing EAOC is the presence of an enhancing mural nodule in a cystic lesion that shows high signal intensity on T1-weighted imaging (T1WI). Additional indicators of malignancy include restricted diffusion on diffusion-weighted imaging (DWI) and loss of the shading sign on T2-weighted imaging (T2WI). Approximately 25% of endometriosis-associated malignancies are extraovarian, most commonly arising in the rectovaginal space and rectosigmoid colon. MRI characteristics suggesting malignant transformation of extraovarian endometriosis include the development of solid masses in the cul-de-sac, pelvic floor or ligaments, characterized by intermediate signal intensity on T2WI, obvious enhancement on post-contrast images, restricted diffusion on DWI, and evidence of metastatic spread. A diagnosis of endometriosis-associated malignancy should be considered when MRI reveals such lesions in women with a previous of endometriosis, or when lesions with endometriotic characteristics coexist with malignant those showing malignant features. In conclusion, MRI plays a crucial role in both the diagnosis and differential diagnosis of endometriosis-related gynecological tumors. This review provided an overview of MRI findings associated with endometriosis-related gynecological tumors at different anatomical sites. Understanding the imaging characteristics of malignant transformation is essential for early detection and timely treatment. With continued advances in MRI technology and improved knowledge of non-invasive diagnostic approaches, the overall accuracy of detecting endometriosis-associated malignancies is expected improved significantly.

Objective: To investigate the conventional magnetic resonance imaging (MRI) findings of borderline ovarian tumor (BOT) and to improve the diagnostic accuracy preoperatively. Methods: The clinical and conventional MRI data of patients confirmed by surgery and pathology at Fudan University Shanghai Cancer Center and Obstetrics & Gynecology Hospital, Fudan University from February 2011 to January 2025 were retrospectively analyzed. Conventional MRI features of the tumors were evaluated including size, location, mass characteristics, loculi, signal intensity of the cystic fluid, the signal intensity of solid component on T2-weighted imaging (T2WI) and diffusion-weighted imaging (DWI), enhancement, ascites, peritoneal seedings, lymph-node involvement and the display of ipsilateral normal ovarian. The differences of these conventional MRI features between serous BOT (SBOT) and mucinous BOT (MBOT) were also compared. Results: A total of 148 BOT patients were included. BOT occurred predominantly in younger women (mean age 40±15). SBOT and MBOT were the two commonest histological subtypes, accounting for 55.4% (82/148) and 29.1% (43/148), respectively. Among the 82 cases of SBOT, 41.5% (34/82) showed bilateral ovarian masses, and 116 masses were found on conventional MRI. Most showed mainly cystic masses (47.4%, 55/116), followed by solid masses (31.0%, 36/116) and mixed cystic-solid masses (21.6%, 25/116). Of all cystic and mixed cystic-solid masses, most were unilocular (75.0%, 60/80) and the signal of cystic fluid was uniformity. The solid component in all mass appeared as papillary projections, most showed a heterogeneous hyper-intensity on T2WI (83.6%, 97/116) and hypo-/iso-intensity on DWI (65.5%, 76/116). The branching papillary projections with hypo-intense fibrous stalks on T2WI were observed in 45 of all masses (38.8%, 45/116) and all 36 solid masses displayed this sign. The ipsilateral normal ovarian tissues were seen in 76.7% (89/116) masses. All patients with MBOT were unilateral cystic tumors with a bulky volume (100.0%, 43/43). Honeycomb loculi were seen 79.1% (34/43) masses. The signal of cystic fluid was heterogeneous, with a mixed low-high signal on T1WI (83.7%, 36/43) and a contained hypo-intensity (41.9%, 18/43) or a heterogeneous iso- to hyper-intensity on T2WI (37.2% 16/43). Most solid components appeared as irregularly thickened septa (81.4%, 35/43), which showed mild-to-moderate enhancement. And 10 masses (23.3%, 10/43) contained “pseudo-solid” areas- T2WI low signal zones rich in mucinous microcysts. Other less common type of BOT did not show some characteristic features, most of which were similar to SBOT. Tumor size, location, mass characteristics, loculi, signal intensity of the cystic fluid, the signal intensity of solid component on T2WI and DWI, enhancement, peritoneal seedings, and the display of ipsilateral normal ovarian differed significantly between SBOTs and MBOTs (all P<0.001). Conclusion: On conventional MRI, BOT have some characteristic features, which are helpful for the accurate diagnosis preoperatively, and can also effectively distinguish SBOT from MBOT.

Objective: To explore the value of computed tomography (CT) and magnetic resonance imaging (MRI) in the diagnosis of cystic-solid ovarian masses in children. Methods: A retrospective analysis of pediatric patients with ovarian cystic-solid masses confirmed by pathological examination at the Xinhua Hospital Affiliated to Shanghai Jiao Tong University School of Medicine from January 2012 to April 2025 was performed. Tumor characteristics on CT and MRI images of pediatric patients, including side, size, calcification, fat content, hemorrhage, diffusion-weighted imaging (DWI) signal, apparent diffusion coefficient (ADC), and enhancement degree were analyzed. Enhanced images were observed for ovarian vascular pedicle signs, as well as enlarged abdominal and pelvic lymph nodes, pelvic effusion, and hydronephrosis. Results: A total of 22 pediatric patients with ovarian cystic-solid masses were included in the study, aged 3 to 12 years. Among them, 5 cases were benign tumors, 6 cases were borderline tumors, and 11 cases were malignant tumors; 14 cases were germ cell tumors, 7 cases were sex cord-stromal tumors, and 1 case was an epithelial tumor. Abdominal and pelvic CT scans were performed alone in 14 cases, pelvic MRI scans were performed alone in 6 cases, and both abdominal and pelvic CT scans and pelvic MRI scans were performed simultaneously in 2 cases. The display rates of the vascular pedicle sign of the affected side of the ovary by enhanced CT and MRI were 86.7% and 71.4%, respectively. The preoperative diagnostic accuracy of CT and MRI for cystic-solid ovarian lesions in children was 75.0%. There was no significant difference in tumor size among benign, borderline and malignant groups (F=0.490, P>0.05). Conclusion: In this study, approximately half of the ovarian cystic and solid lesions in children were malignant tumors. Some tumors exhibited characteristic features on CT and MRI scans. Combining tumor markers or sex hormone levels can improve the accuracy of preoperative diagnosis.

Objective: To develop a dual-center magnetic resonance imaging (MRI) radiomics nomogram for preoperative prediction of risk stratification in endometrial cancer (EC). Methods: Patients who underwent MRI examinations prior to surgery were collected and anlyzed at Shaanxi Provincial Cancer Hospital (Institution 1) and Xi'an Central Hospital (Institution 2) between September 2015 and February 2022 and were confirmed to have EC by postoperative pathological examination. Tumor segmentation was performed manually on axial T2-weighted imaging (T2WI) and apparent diffusion coefficient (ADC) images using 3D Slicer software. Radiomics features were extracted from intratumoral, peritumoral, and combined intratumoral and peritumoral regions. Least absolute shrinkage and selection operator (LASSO) regression was applied for feature selection, and radiomics models were constructed for different sequences (ADC and T2WI) and tumor regions. Diagnostic performance was evaluated via area under the curve (AUC), and the optimal model was selected. Clinical parameters, tumor volume, tumor size, maximum anteroposterior tumor diameter on sagittal T2WI (APsag), and tumor to uterine area ratio (TAR) were screened and integrated with radiomics features using logistic regression (LR) to build a clinical model. A combined MRI radiomics nomogram was developed by integrating the clinical and radiomics models. Model performance was assessed through receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis. Results: A total of 293 patients with EC pathologically confirmed from two institutions who underwent preoperative 1.5 T MRI were retrospectively enrolled. Institution 1 contributed 233 patients, randomly divided into a training cohort (n=163) and internal validation cohort (n=70) at a 7∶3 ratio. Institution 2 provided 60 patients as an external validation cohort. Finally, 9 radiomics features significantly associated with EC risk stratification were identified. The radiomics model demonstrated high diagnostic efficacy across all cohorts (AUC_training=0.926, AUC_{internal validation}=0.912, AUC_{external validation}=0.869). Incorporation of clinical parameters into the nomogram further enhanced performance (AUC_training=0.949, AUC_{internal validation}=0.951, AUC_{external validation}=0.932). Conclusion: The MRI-based radiomics nomogram exhibits robust predictive accuracy for preoperative EC risk stratification, offering valuable guidance for clinical decision-making.

Objective: To analyze and summarise the clinical and ultrasound manifestations of low-grade endometrial stromal sarcoma (LG-ESS), and to explore the clinical value of ultrasound detection and diagnosis of LG-ESS. Methods: Clinical data and ultrasound images of patients of LG-ESS diagnosed by surgical pathology at Shanghai Changning Maternity and Infant Health Hospital, Maternity and Infant Health Hospital Affiliated to East Normal University, were retrospectively collected from January 2012 to December 2024. Ultrasound characteristics such as tumour location, size, borders and internal echoes were analyzed. Results: A total of 12 LG-ESS patients were enrolled, and 66.7% of the patients in this group had clinical manifestations such as irregular vaginal bleeding and increased menstrual volume, and carbohydrate antigen (CA)125 was not significantly increased. The 12 cases of LG-ESS can be classified into 4 types according to the locations of the tumours, including uterine cavity type (n=4), muscular wall type (n=5), uterine cavity and muscular wall type (n=2), and extrauterine type (n=1). The mean maximum tumour diameter of the tumors was (56.3±23.2) mm (16 to 88 mm). The tumour boundaries were indistinct in 5 cases, unclear in 6 cases and clear in 1 case (extrauterine type). The internal echo of the tumours could be classified into 4 types, solid type (n=7), diffuse type (n=1), mixed type (n=3) and cystic type (n=1). Both the ultrasound preoperative detection rate and accuracy of localization were 100.0% (12/12). The coincidence rate of preoperative ultrasound diagnosis and pathological diagnosis was 0 (0/12). Conclusion: LG-ESS sonograms are variable in presentation and preoperative ultrasound diagnosing has some limitations, but ultrasound is still an important initial screening tool. In middle-aged women with irregular vaginal bleeding, when there is a relatively large solid or mixed mass in the uterus, the possibility of LG-ESS should also be considered when considering uterine common diseases such as uterine fibroids or myoadenosis or cystic degeneration of fibroids.

Objective: To evaluate the clinical utility of a combined model incorporating radiomics features from B-mode ultrasound and blood flow characteristics from color Doppler ultrasound for predicting sentinel lymph node (SLN) metastasis in breast cancer. Methods: This study retrospectively involved breast cancer patients who underwent ultrasound examinations in Shanghai Pudong New Area People's Hospital between October 2022 and December 2024. Color Doppler ultrasound was used to assess intratumoral blood flow signals, analyzing both overall and regional blood flow characteristics of the lesion. Three SLN metastasis prediction models were developed using a support vector machine (SVM): a model based solely on B-mode ultrasound radiomics features (US model), a model based solely on color Doppler ultrasound blood flow features (CDUS model), and a combined model integrating both feature sets (COMB model). The diagnostic performance of each model was evaluated using receiver operating characteristic (ROC) curve analysis. Results: A total of 328 breast cancer patients were enrolled and randomly divided into a training set (230 cases) and a test set (98 cases) in a 7∶3 ratio. In both the training and test sets, the overall blood flow ratio in the SLN metastasis group (2.4 ± 0.6 and 2.1±0.9) was significantly higher than that in the non-metastasis group (1.3 ± 0.7 and 1.2 ± 0.8) (both P<0.001). ROC curve analysis showed that the COMB model achieved the highest area under curve (AUC) in both the training (0.913, 95% CI 0.869-0.946) and test sets (0.811, 95% CI 0.721-0.901). The US model ranked second, with AUCs of 0.796 (95% CI 0.738-0.846) and 0.757 (95% CI 0.660-0.838), while the CDUS model demonstrated the lowest performance, with AUCs of 0.704 (95% CI 0.614-0.795) and 0.655 (95% CI 0.545-0.765). Conclusion: Color Doppler ultrasound effectively assesses the overall blood flow ratio of breast cancer lesions, which is closely associated with SLN metastasis. The COMB model, integrating B-mode ultrasound radiomics features and color Doppler blood flow characteristics, significantly improves the accuracy of SLN metastasis prediction, providing valuable insights for the precise diagnosis and treatment of breast cancer.

Objective: To evaluate the diagnostic efficacy of multiparameter magnetic resonance imaging (mpMRI) fusion with transrectal ultrasound (TRUS) navigation combined with strain elastography (SE) guided prostate biopsy for the detection of prostate cancer. Methods: From January 1, 2022 to April 1, 2024, patients with suspected prostate cancer were collected and analyzed in the Second Affiliated Hospital of Soochow University. All patients underwent mpMRI, SE, and TRUS examinations. Following the identification of suspected prostate cancer lesions, both a standard prostate biopsy consisting of ten systematic needle punctures and a targeted biopsy guided by fusion software combining mpMRI-TRUS images were performed. The diagnostic efficacy of each individual biopsy method as well as their combined application in differentiating benign from malignant prostate lesions was evaluated. Results: A total of 96 suspected lesions were identified in 66 patients suspected of prostate cancer. The positive rate of systematic puncture was found to be 63.64%. The accuracy of the mpMRI group was determined to be 71.88%, while the SE group had an accuracy rate of 66.67%. In comparison, the TRUS group showed an accuracy rate of 60.42%. However, when combining mpMRI fusion navigation with SE imaging, a higher accuracy rate of 86.46% was achieved. Comparative analysis of the diagnostic performance among SE, mpMRI, and combined imaging using receiver operating characteristic (ROC) curve analysis showed that the area under the curve (AUC) for the combined imaging group was 0.717, which was higher than that of the SE and mpMRI groups. The difference was statistically significant. Furthermore, in terms of sensitivity and negative predictive value (NPV), the combined imaging group exhibited values at 94.03% and 69.23%, respectively. In systemic biopsy, a total number of 660 needles were punctured resulting in a single needle positive rate at 27.73%. Conversely, within the combined imaging group where a total number of only 200 needles were used for puncture procedures, there was a significantly higher single needle positive rate recorded at 81.50% (P<0.001). Conclusion: Targeted prostate biopsy guided by TRUS-fused mpMRI combined with SE demonstrates a significantly higher prostate cancer detection rate compared to conventional biopsy techniques, while simultaneously reducing the number of biopsy cores required. This approach offers notable clinical value in the diagnosis of prostate cancer.

Objective: To explore the effect of deep learning-based reconstruction (DLR) on the quality and diagnostic value of T2-weighted imaging FatSat (T2WIFS) sequences. Methods: A retrospective analysis of clinical data from patients with suspected prostate cancer (PCa) at Fudan University Shanghai Cancer Center from March to December 2024 was conducted. Patients were divided into a PCa group and a benign prostatic hyperplasia (BPH) group based on needle biopsy and postoperative pathological examination. Scanning suppression sequences included conventional T2 FatSat (T2FSC) and deep learning-based reconstruction T2 FatSat (T2FSDL) of DLR. The overall image quality and image artifacts of prostate imaging were evaluated by two physicians on a five-point scale. The objective evaluation was the signal intensity (S) and standard deviation (SD) of the prostate images of T2FSDL and T2FSC, where the SD was regarded noise (N), and the signal noise ratio (SNR) was calculated. T test was performed for normally distributed data, and Wilcoxon rank sum test was performed for non-normally distributed data. The subjective scores and objective indexes of T2FSDL images and T2FSC images were compared and analyzed. Weighted-Kappa test was used to compare the inter-group and intra-group subjective rating consistency. The prostate lesions in T2FSDL images and T2FSC images were scored by PI-RADS by two film readers using double-blind method. Receiver operating characteristic (ROC) curve was drawn based on pathological results. The area under curve (AUC) was calculated to analyze the diagnostic value of each image for prostate cancer. Results: A total of 116 patients with suspected PCa were included in the retrospective study of this experiment, including 68 patients with malignant PCa and 48 patients with BPH. The subjective scores and objective measurement data of the two groups of images (T2FSDL and T2FSC sequences) were in good agreement between the two physicians (Kappa>0.8). In terms of subjective scores, the overall quality scores of T2FSC and T2FSDL images were (4.04±0.68) and (4.53±0.54), with statistical significance(P<0.01); pseudo-film ratings were (4.44±0.68) and (4.35±0.66), with no statistical significance (P=0.34). In terms of objective evaluation, the SD of T2FSDL images (0.65±0.19) was lower than that of T2FSC images (1.09±0.24), with statistical significance(P<0.01), and the SNR of T2FDL images (95.61±14.25) was higher than that of T2FSC images (56.48±9.72), with statistical significance (P<0.01). In terms of diagnostic value of prostate cancer, the AUC corresponding to T2FSDL images (0.866) was greater than that corresponding to T2FSC images (0.819), with statistical significance (P<0.01). The scanning time of T2FSDL sequence was 100 s, which was significantly faster than that of T2FSC sequence 150 s. Conclusion: T2FSDL technology based on deep learning reconstruction can effectively improve the quality of MRI and the diagnostic value of T2FSDL images is also higher than that of T2FSC images, and can significantly shorten the scanning time and optimize the efficiency of prostate scanning, which has a good clinical application prospect.

Objective: To build radiomics models based on the features from computed tomography (CT) imaging modality and investigate the predictive efficacy for immunotherapy in patients with non-small cell lung cancer (NSCLC). Methods: CT imaging data and clinical records in stage III-IV NSCLC patients treated with programmed death-1 (PD-1)/programmed death-ligand 1 (PD-L1) inhibitors monotherapy at Huadong Hospital Affiliated to Fudan University from June 2018 to December 2024 were retrospective analyzed. Radiomics features were extracted from the CT image modalities for analysis. The least absolute shrinkage and selection operator (LASSO) regression analysis method based on R language machine learning algorithm with 10-fold cross-validation was used to construct the radiomics prediction model. The area under the receiver operating characteristic curve (AUC) was used to evaluate the prediction model. Results: A total of 141 patients were included, and 1 218 radiomic feature parameters were extracted. The radiomics signatures were constituted by nine selective features (1 shape feature, 2 first-order features, and 6 texture features), showed good discrimination for the effect of immunotherapy in patients with NSCLC, with area under curve (AUC), sensitivity, and specificity of 0.912 (95% CI 0.837-0.960), 0.918, and 0.837, respectively in training group and 0.878 (95% CI 0.742-0.958), 0.720, and 0.944, respectively in validation group. The decision curve analysis (DCA) showed that using this model can improve clinical decision-making benefits when the clinical decision threshold is less than 0.4. Conclusion: The CT radiomics model holds substantial value in predicting the efficacy of immunotherapy for NSCLC. It not only aids in identifying patients who are responsive to immunotherapy but also facilitates the optimization of treatment strategies, thereby achieving personalized precision medicine.

Objective: To investigate the necessity of incorporating peripheral immune cells into machine learning models for predicting microvascular invasion (MVI) in hepatocellular carcinoma (HCC), and to interpret the best-performing machine learning model while exploring the radiomics features with the greatest contribution. Methods: Patients from Yangpu Campus, Third Affiliated Hospital of Naval Medical University (training set), Jiading Campus, Third Affiliated Hospital of Naval Medical University (testing set), and Suzhou Municipal Hospital (validation set) between January 2019 and December 2021 were retrospectively included. Five machine learning methods were used to construct radiomics signatures and clinical signatures. The optimal model was identified through fusion and visualized via nomograms. Finally, the radiomics features with the greatest contribution to the model were explored. Results: A total of 276 patients were included, with 189 in the training set, 47 in the testing set, and 40 in the validation set. The extreme gradient boosting (XGBoost) classifier, as the best-performing model, achieved areas under the curve (AUC) of 0.962 (95% CI 0.939-0.989) in the training set, 0.808 (95% CI 0.682-0.934) in the testing set, and 0.816 (95% CI 0.774-0.858) in the validation set. When excluding peripheral immune cells in the comparison set, XGBoost's AUC was 0.925 (95% CI 0.890-0.960). DeLong test showed a significant difference between the training set and comparison set (Z=-3.083, P=0.002). Additionally, the radiomics model was interpreted through histograms and decision trees, identifying “original_shape_Sphericity” as the most important radiomics feature. Conclusion: Among the five machine learning models for predicting MVI, XGBoost performed best, and incorporating peripheral immune cells significantly enhanced model performance. Among radiomics features, shape features were particularly important for predicting MVI.

Objective: To explore the application value of magnetic resonance elastography (MRE) in differentiating benign and malignant focal liver lesions and to analyze its potential in predicting the prognosis of malignant tumors. Methods: A retrospective analysis was conducted on patients with hepatic focal lesions who underwent routine magnetic resonance imaging (MRI) and MRE scans at Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine from December 2021 to October 2024. The cases included hepatic hemangiomas, hepatocellular carcinoma, liver metastases, and cholangiocarcinoma. 2D MRE scans were performed on a 3.0 T MRI scanner to obtain tumor elasticity maps and measure lesion elasticity values. Univariate analysis of variance was used to compare the differences in MRE elasticity values among various lesions and to assess the diagnostic efficacy of MRE in benign and malignant lesions. The optimal cutoff value of MRE elasticity for distinguishing benign and malignant lesions was determined through receiver operating characteristic (ROC) curve analysis, and diagnostic performance metrics such as the area under the curve (AUC) were calculated. Paired sample t-tests were used to compare elasticity value differences before and after treatment. Results: This study included 172 patients with localized liver lesions, including 50 cases of hepatic hemangioma, 59 cases of hepatocellular carcinoma, 51 cases of liver metastasis, and 12 cases of cholangiocarcinoma. Hepatocellular carcinoma, liver metastases and cholangiocarcinoma exhibited elasticity values of (8.97±5.33) kPa, (7.52±4.64) kPa, and (8.74±4.82) kPa, respectively; whereas hepatic hemangiomas had notable lower elasticity values of (3.10±1.30) kPa, with a statistically significant difference between the malignant tumors group and the hepatic hemangiomas group (F=66.080，P<0.001). The MRE elasticity value's AUC for distinguishing between benign and malignant lesions was 0.870, with sensitivities, specificities, positive predictive values, and negative predictive values at 75.4%, 84.0%, 92.0%, and 58.3%, respectively, and the optimal cutoff value was 4.15 kPa. In the analysis of 30 malignant tumor patients from different efficacy groups, the disease progression group showed a marked increase in elasticity values from (5.98±2.48) kPa before treatment to (10.74±3.83) kPa after treatment, with a statistically significant difference (t=-5.134, P<0.001). And the partial remission group showed elasticity values of (4.31±1.88) kPa before and (3.06±1.43) kPa after treatment, with a statistical difference (t=4.411，P=0.003). In contrast, the stable disease group showed values of (10.64±7.71) kPa before and (10.32±7.80) kPa after treatment, with no significant statistical differences observed for either group (t=1.209，P=0.258). Conclusion: MRE, as a non-invasive technique, can effectively differentiate between benign and malignant liver lesions and shows potential value in assessing the prognosis of malignant tumors, providing important diagnostic and therapeutic guidance for clinical practice.

Objective: To analyze the imaging characteristics of computed tomography (CT) and magnetic resonance imaging (MRI) multimodal imaging techniques in children with hepatoblastoma (HB), and to explore the application value of CT and MRI parameters in HB diagnosis and pre-treatment extent of tumor (PRETEXT) staging. Methods: Children with HB diagnosed at Xuzhou Children's Hospital from January 1, 2019 to December 31, 2023 were selected and analyzed. The children were categorized into two groups according to the pathological type of the surgical and puncture biopsy findings: complete epithelial and mixed epithelial-mesenchymal. The differences in birth history, general characteristics of the lesion, internal characteristics of the lesion, mode of enhancement, and perihepatic effusion between the 2 groups were compared. Children with HB were divided into two groups according to PRETEXT staging: stage Ⅰ-Ⅱ and stage Ⅲ-Ⅳ. The differences in CT values of tumors between plain CT scans and enhanced CT scans of lesions, as well as the apparent diffusion coefficient (ADC) of lesions on MRI scans were compared. Logistic multivariable regression analysis were used to identify predictive factors for PRETEXT staging in HB patients. The predictive value of PRETEXT staging in HB patients was analyzed using receiver operating characteristic (ROC) curves. Results: Sixty HB pediatric patients were included, with 30 cases each of complete epithelial type and mixed epithelial-mesenchymal type. There was no significant difference in the birth histories of children with HB between the two groups (P>0.05). However, the proportions of intra-tumor hemorrhage, tumor calcification, and perihepatic effusion were higher in the mixed type of HB than in the complete type of HB (P<0.05). Logistic regression analysis showed that intra-tumor hemorrhage, tumor calcification, and perihepatic effusion were of value in determining the pathological type of children with HB (OR=4.75, 2.77, 0.92, P<0.05); children with stage Ⅲ-Ⅳ HB had higher plain CT values [(46.95±5.28) HU], enhanced scan CT values in the arterial and delayed stages [(98.04±21.91) HU, 92.64±14.33) HU], and ADC values [(1.22±0.15)×10^-3 mm²/s] than those in stage Ⅰ-Ⅱ (P<0.05), and enhanced scan CT values in the venous stage [(94.12±20.28) HU] were lower than those of stage I~Ⅱ (P<0.05). Logistic multifactorial regression analysis showed that high plain CT values, high CT values on arterial phase enhancement scans, and high ADC values were independent influences on PERTEXT staging (OR=1.14, 1.04, and 15.97, P<0.05). ROC curves analysis showed that the area under curve (AUC) of the three combined tests for PERTEXT staging was 0.890 higher than that of 0.663, 0.718, and 0.791 (P<0.05). ROC curve analysis showed that the AUC of the three combined detections for PRETEXT staging was 0.890, which was higher than the 0.663, 0.718 and 0.791 predicted separately (P<0.05). Conclusion CT and MRI multimodal imaging technology can accurately predict the pathological staging and PRETEXT staging of HB through the combined analysis of quantitative parameters (plain CT value, arterial phase CT value, ADC value) and morphologic features (calcification, hemorrhage, and periportal hepatic effusion), which can provide a key basis for the selection of preoperative chemotherapy regimen, planning of surgical scope, and prognostic assessment, and has a good clinical translational value.

The early diagnosis and timely intervention of prostate cancer (PCa) are critical for improving patient prognosis. Ultrasound technology, with its economic affordability, convenience, and real-time intraoperative imaging capabilities, has emerged as a vital tool in the diagnosis and treatment of PCa. In recent years, the application of artificial intelligence (AI) in ultrasound-based PCa diagnosis and treatment has demonstrated remarkable progress. This review examined the advancements of ultrasound AI in the field of PCa from three perspectives: traditional machine learning, deep learning, and reinforcement learning. Machine learning, through feature engineering and model training, has optimized the segmentation accuracy of ultrasound images. Furthermore, by integrating multimodal ultrasound imaging, clinical and proteomic features, machine learning models have enhanced diagnostic accuracy for PCa and improved the ability to differentiate metastatic PCa. Deep learning, leveraging its end-to-end learning capability, not only effectively identifies PCa lesions but also enables precise grading of these lesions. Additionally, in brachytherapy for PCa, deep learning has accurately delineated the clinical target volume of the prostate from ultrasound images and efficiently automated the reconstruction of catheters and treatment needles, thereby reducing human operational errors and procedure time. Reinforcement learning, through its trial-and-error interaction and cumulative reward mechanism, has enabled automated quality assessment of ultrasound images to select high-quality images for AI model training and thereby alleviating the workload of manual image annotation by clinicians. Moreover, reinforcement learning has improved the accuracy and reliability of PCa biopsies by reducing intraoperative sampling errors through adaptive intraoperative planning. Finally, this review analyzed the current challenges faced in the clinical application of ultrasound AI for PCa. Future efforts should focus on establishing multi-center, high-quality shared datasets, developing ultrasound AI models with strong interpretability, safety, and clinical practicality, and formulating standardized regulatory strategies. These steps are essential to ensure that ultrasound AI systems can effectively serve the precise and personalized diagnosis and treatment of PCa patients.