Objective: To assess the feasibility and accuracy of a deep learning model based on two-dimensional ultrasound images for preoperative prediction of lymphovascular invasion (LVI) in breast cancer, and compare its results with traditional interpretations by ultrasound physicians, providing novel imaging insights to support personalized treatment decisions and precision medicine. Methods: A retrospective analysis was conducted on patients diagnosed with breast cancer through postoperative pathology, who underwent breast surgery in Sixth People’s Hospital Affiliated to Medical College of Shanghai Jiao Tong University between January 2020 and December 2023. Standardized grayscale ultrasound images of lesions were processed, and features were extracted using a convolutional neural network (CNN)-based deep learning model to predict LVI. The model’s performance was evaluated using accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). Comparisons were made with the diagnostic performance of ultrasound physicians with varying levels of experience, and the impact of the deep learning model on improving physician diagnostics was analyzed. Results: A total of 232 patients were included, with a total of 232 lesions. The dataset was divided into a training set (185 cases) and a validation set (47 cases) in an 8∶2 ratio. Among the 232 patients, 102 cases (43.97%) were confirmed to have LVI by postoperative pathology. The deep learning model (support vector machine) achieved AUCs of 0.917 (95% CI 0.877-0.956) and 0.863 (95% CI 0.750-0.975) in the training and validation sets, respectively. For the validation set, the accuracy, sensitivity, and specificity were 83.0%, 85.7%, and 80.8%, respectively. In comparison, ultrasound physicians with 5-10 years and 10-15 years of experience achieved AUCs of 0.623 and 0.709, respectively, which were significantly lower than the deep learning model (P<0.05). When assisted by the deep learning model, the AUC for the physician with 10-15 years of experience increased to 0.914, with corresponding improvements in accuracy (91.5%), sensitivity (90.5%), and specificity (92.3%). Conclusion: The deep learning model based on two-dimensional ultrasound images demonstrated superior accuracy and reliability in the preoperative prediction of LVI in breast cancer. It significantly outperformed traditional diagnostic approaches by ultrasound physicians and showed potential as a clinical auxiliary tool to improve the precision of preoperative assessments of LVI, supporting personalized patient treatment planning.
Objective: To investigate the application value of deep learning ultrasound radiomics in predicting axillary lymph node (ALN) metastasis of breast cancer. Methods: The ultrasound images of breast cancer patients pathologically confirmed were retrospectively analyzed in Gongli Hospital, Shanghai Pudong New Area, from January 2021 to December 2023. Based on whether ALN had metastasized, patients were divided into two groups: those without ALN metastasis and those with ALN metastasis. The research focused on correlating ultrasound characteristics of primary breast cancer lesions with ALN metastasis and evaluating their predictive efficacy. The dataset was randomly split into training and testing sets at an 8∶2 ratio. Nine deep learning models, ResNet50, EfficientNet, MobileNetV3, DenseNet121, DenseNet201,Vision Transforme, VGG16, MobileViT, and Mamba Transformer, were used to predict ALN metastasis. Through five-fold cross-validation, the best-performing model was selected, and decision curve analysis (DCA) was conducted to assess the clinical net benefit of each model. The study also compared the predictive performance of deep learning models against traditional ultrasound features in identifying ALN metastasis in breast cancer patients. Results: A total of 324 breast cancer patients were included in the study, with a total of 324 breast lesions. Among them, 198 cases had no ALN metastasis, and 126 cases had ALN metastasis. Univariate analysis revealed statistically significant differences (P<0.05) between the non-ALN metastasis and ALN metastasis groups in terms of primary lesion characteristics, including size, shape, orientation, margin, calcification, echogenic halo, spiculation, and lobulation. Multivariate logistic regression identified larger lesion size, non-parallel orientation, presence of an echogenic halo, and spiculation as independent risk factors for ALN metastasis. The combined diagnostic performance of these four features yielded an area under curve (AUC) of 0.805. Among the nine deep learning models evaluated, DenseNet201 demonstrated the highest performance, with AUCs of 0.964 (training set) and 0.861 (testing set). The deep learning models outperformed traditional ultrasound features in predicting ALN metastasis. DCA of DenseNet model indicated a significant net benefits within a risk threshold range of 0.170 6 to 0.605 2. Conclusion: Deep learning ultrasound has high clinical value in non-invasive evaluation of axillary lymph node metastasis of breast cancer before surgery, and can provide a basis for the selection of preoperative diagnosis and treatment.
Objective: To investigate the predictive value of multimodal ultrasound features combined with serum carcinoembryonic antigen (CEA) and cytokeratin (CK)19 levels for axillary lymph node metastasis in breast cancer. Methods: A retrospective analysis was conducted on breast cancer patients in The First Hospital of Qinhuangdao from February 2023 to August 2024, assessing their multimodal ultrasound features and serum CEA and CK19 levels at initial diagnosis. According to the pathological results of axillary lymph nodes, patients were categorized into a metastasis-negative group and a metastasis-positive group. Comparative analyses were performed between the two groups regarding the maximum diameter of the primary breast lesion, margin characteristics, presence of microcalcifications, orientation (parallel or non-parallel based on the length-to-width ratio), Adler classification of blood flow imaging, and shear wave elastography parameters (Emax and Emin values). Additionally, serum CEA and CK19 levels were evaluated. The predictive value of combining multimodal ultrasound features with serum CEA and CK19 levels for axillary lymph node metastasis was explored. Results: This study included a total of 136 breast cancer patients, including 63 cases in the axillary lymph node metastasis-negative group and 73 cases in the metastasis-positive group. Univariate analysis and logistic regression identified the presence of microcalcifications in the primary breast lesion, Adler grade Ⅱ-Ⅲ blood flow, elevated Emax values, and increased serum CEA levels as independent predictors of axillary lymph node metastasis in breast cancer. A predictive model was constructed based on these four factors, achieving an area under the curve (AUC) of 0.910 (95% CI 0.862-0.957). Conclusion: The presence of microcalcifications in the primary breast lesion, Adler grade Ⅱ-Ⅲ blood flow, elevated Emax values, and increased serum CEA levels indicate a higher likelihood of axillary lymph node metastasis in breast cancer. A logistic regression model constructed with these four predictive factors demonstrates potential for predicting axillary lymph node metastasis in breast cancer.
Objective: To examine the imaging characteristics of mucinous breast carcinoma (MBC) in multi-parametric ultrasonography (mpUS), including conventional ultrasound, shear wave elastography (SWE), and contrast-enhanced ultrasound (CEUS), and evaluate their diagnostic efficacy for MBC. Methods: Patients admitted to Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine for consultation regarding breast lesions between January 2017 and April 2023 were retrospectively included in the study, with post-surgical resection pathology serving as the gold standard for diagnosis. Patients with fibroadenomas (FA) were randomly selected as controls from the same period of time in a 1∶1.5 ratio according to the random number table method. The imaging features were analyzed and classified according to the Breast Imaging Reporting and Data System (BI-RADS). Logistic regression was employed to identify valid diagnostic indicators, construct a prediction model, and generate the receiver operating characteristic curves for the subjects, allowing for the calculation of the area under curve (AUC). The AUC was used to assess the diagnostic efficacy. Results: The final sample included 40 lesions from 40 patients with MBC, comprising 16 cases of pure MBC (PMBC), 24 cases of mixed MBC (MMBC), and 60 FA lesions from 60 patients with FA. On conventional ultrasound, 50.0% (8/16) of PMBC were classified as BI-RADS category 3 and 4A, and 70.8% (17/24) of MMBC were classified as BI-RADS≥4B. In SWE, PMBC and MMBC were more frequently characterized by inhomogeneous hardness than FA (62.0% and 83.3% vs 25.0%), and a score of 3-4 was an independent predictor for distinguishing both from FA. In CEUS, the presence of cystic non-enhancing areas within the lesion and increased trophoblastic vessels at the margins were identified as independent risk factors for differentiating PMBC from FA. Compared with MMPC, mpUS demonstrated a significant improvement in the differential diagnosis of PMBC from FA when used in conjunction with conventional ultrasound (AUC: 0.949 vs 0.858, P<0.05). Conclusion: MBC, particularly PMBC, is frequently underestimated due to its resemblance to fibroadenomas on conventional sonograms. The differential diagnosis of MBC and FA can be significantly enhanced by mpUS in comparison to conventional ultrasound alone.
Objective: To explore the application value of gastric oral contrast-enhanced ultrasound and gastroscopy in early diagnosis of gastric diseases in community hospitals. Methods: Patients with gastric diseases admitted to Pujiang Community Health Service Center from June 2021 to August 2024 were selected. All patients underwent gastroscopy and oral contrast-enhanced ultrasound, with pathological diagnosis as the gold standard. The diagnostic accuracy of gastric oral contrast-enhanced ultrasound, gastroscopy, and combined examination with pathological diagnosis results were evaluated. And the thickness of the gastric wall for benign and malignant gastric diseases was compared. Results: A total of 268 patients were included. Pathological diagnosis detected 183 positive cases, including 30 malignant cases. There were 151 positive cases detected by gastric oral contrast-enhanced ultrasound, including 22 malignant cases. There were 160 positive cases detected by gastroscopy, including 22 malignant cases; 175 positive cases were detected by joint examination, including 25 malignant cases. The diagnostic accuracy of combined examination for gastritis was 96.91%, which was higher than the 83.51% of gastric oral contrast-enhanced ultrasound (χ2=10.533，P<0.05). The sensitivity of combined examination in diagnosing benign and malignant gastric lesions was 80.00%, which was higher than the 50.00% and 53.00% of gastric oral contrast-enhanced ultrasound and gastroscopy examinations, respectively. The accuracy of the combined examination was 97.39%, which was higher than the 91.79% and 92.54% of gastric oral contrast-enhanced ultrasound and gastroscopy examinations, respectively (χ2=6.826, 8.650; P<0.05). The thickness of the gastric wall in patients with malignant gastric diseases was greater than that in patients with benign gastric diseases (t=20.818, P<0.05). Conclusion: The combined examination of gastric oral contrast-enhanced ultrasound and gastroscopy could improve the detection of gastric diseases and enhance the diagnostic value of benign and malignant gastric diseases, while the thickness of the gastric wall in malignant lesions was greater than that in benign lesions.
Objective: To construct a nomogram model for predicting death in cervical cancer patients within 5 years based on color Doppler ultrasound parameters and serological indicators, and to evaluate the discrimination and consistency of the model. Methods: From October 2020 to December 2024, patients with cervical cancer who received radical cervical resection in Longgang District People’s Hospital in Shenzhen were regarded as the research objects. All patients underwent preoperative color Doppler ultrasound. Serological indicators were collected. Multivariate COX regression analysis of the risk factors for death within 5 years in cervical cancer patients. R software was used to build a nomogram model to predict the 5-year mortality risk of cervical cancer patients, and the receiver operating characteristic (ROC) curve and calibration curve were used to verify the discrimination and consistency of the nomogram model. Results: A total of 500 cervical cancer patients were included, and the cervical cancer patients were grouped into a modeling set of 300 cases and a validation set of 200 cases in a ratio of 6∶4. The resistance index (RI) and pulsatility index (PI) of the death group were lower than those of the survival group, and the proportions of abundant microvascular blood flow, SCCA≥1.5 ng/mL, and CA125≥35 kU/L in the lesion were higher than those of the survival group (P<0.05). COX regression analysis showed that RI index, abundant intralesional microvascular blood flow, SCCA≥1.5 ng/mL, CA125≥35 kU/L were risk factors affecting the prognosis of cervical cancer patients (P<0.05). Based on risk factors, R software was used to establish a nomogram model to predict the 5-year mortality risk of cervical cancer patients, and the Hosmer-Lemeshow goodness of fit test showed that the modeling set χ2=7.629, P=0.471, the validation set χ2=9.051, P=0.338. The area under the ROC curve of the modeling set was 0.841. The area under curve of the validation set was 0.822. Conclusion: The nomogram model constructed in this study to predict the 5-year mortality risk of cervical cancer patients has good discrimination and consistency.
Objective: To investigate the value of machine-learning-based radiomics in predicting disease-free survival (DFS) and overall survival (OS) after concurrent chemoradiotherapy in patients with locally advanced cervical cancer. Methods: Three-dimensional radiomics parameters of the primary lesion and its surrounding 5 cm region in T2-weighted imaging (T2WI) sequences of all patients were measured. Six machine learning methods were used to construct the optimal radiomics model and to analyse its incremental value for existing clinical markers. Results: Data of 632 patients with locally advanced cervical cancer who underwent concurrent chemoradiotherapy with continuous follow-up in two centres were retrospectively analysed. And 552 patients from centre 1 served as the training set and 80 patients from centre 2 served as the validation set. In the prediction of DFS, the combined tumor and peritumor randomised survival forest model showed the best predictive efficacy, with 1-year, 3-year and 5-year area under curve (AUC) of 0.955, 0.906, 0.970, and 0.781, 0.885, 0.836 in the training and validation sets, respectively. In the prediction of OS, the combined tumor and peritumor Deepsurv model showed the best predictive efficacy, with 1-year, 3-year and 5-year AUC of 0.977, 0.939, 0.933, and 0.846, 0.875, 0.808 in the training and validation sets, respectively. Conclusion: Machine learning-based radiomics model helps to predict DFS and OS after concurrent chemoradiotherapy in cervical cancer patients, and the combination of radiomics and clinical indicators has higher predictive efficacy, which can provide a reliable basis for diagnostic decision-making and prognostic prediction in cervical cancer patients.
Objective: To explore the predictive value of artificial intelligence (AI) software in identifying benign and malignant lung nodules and predicting the pathological types of lung nodules. Methods: Patients with lung nodules confirmed by pathological examination were collected, who underwent high-risk lung nodule screening at the Minhang Branch of Affiliated to Fudan University Cancer Hospital from September 2020 to August 2024. AI software was used to analyze the benignity and malignancy and pathological types of pulmonary nodules, and consistency with pathological results was tested. The diagnostic performance of the AI software was evaluated through the area under the receiver operating characteristic (ROC) curve. Results: A total of 62 patients with pulmonary nodules were included in the study, including 4 cases of inflammatory nodules, 4 cases of carcinoma in situ, 2 cases of atypical adenomatous hyperplasia, 16 cases of microinvasive adenocarcinoma, 32 cases of invasive adenocarcinoma, and 4 cases of squamous cell carcinoma. The sensitivity, specificity, and accuracy of the pulmonary nodule software in diagnosing the benignity and malignancy of pulmonary nodules were 98.28%, 75.00%, and 96.80%, respectively. The area under the ROC curve for diagnosing benign and malignant pulmonary nodules by AI analysis was 0.866. The consistency between the AI software’s predictions of pulmonary nodule pathological types and the pathological results was tested, with a Kappa value of 0.859. Conclusion: AI software based on computed tomography (CT) target scanning can effectively distinguish between benign and malignant lung nodules during lung cancer screening. It provides a reference for predicting the pathological types of lung nodules and is valuable for optimizing clinical surgical procedures and enabling precise management of patients with pulmonary nodules.
Objective: To analyze the computed tomography (CT), high-resolution magnetic resonance imaging (HR-MRI), diffusion-weighted imaging (DWI) manifestations and parameters of postoperative anastomotic recurrence in rectal cancer patients, and to explore their diagnostic value for recurrence. Methods: Patients with suspected anastomotic recurrence after rectal cancer surgery in Ziyang Central Hospital from January 2020 to December 2022 were selected as the study subjects, all of whom underwent CT, HR-MRI, and DWI examinations. The diagnostic value of CT, HR-MRI, DWI imaging findings, parameters [CT value, signal intensity value, contrast noise ratio (CNR), average apparent diffusion coefficient (ADC) during plain scan, arterial phase, venous phase] in the diagnosis of anastomotic recurrence after rectal cancer surgery was analyzed. Results: A total of 103 patients with suspected anastomotic recurrence after surgery for rectal cancer were included in the study. Pathological examination confirmed that 62 cases had recurrence and 41 cases had no recurrence. CT imaging findings of recurrent patients: patients with anastomotic recurrence after rectal cancer surgery have narrowed anastomotic gaps, irregular thickening of the outer edge, and involvement of pelvic tissues and organs; HR-MRI imaging findings: the intestinal wall in the surgical area of patients with anastomotic recurrence after rectal cancer surgery showed varying degrees of thickening, with slightly longer signals on T1-weighted imaging (T1WI) and longer signals on T2-weighted imaging (T2WI). The enhanced scan showed significant enhancement. Patients with no anastomotic recurrence after rectal cancer surgery showed only mild enhancement in the T1WI signal area, which was fibrous scar tissue; DWI imaging findings: patients with anastomotic recurrence after rectal cancer surgery showed slightly high or high signal intensity on DWI and low signal intensity on ADC maps. Patients with no anastomotic recurrence after rectal cancer surgery showed low signal intensity or slightly high signal intensity on DWI and high signal intensity on ADC maps. The CT value, signal intensity value, and CNR of recurrent patients during the arterial and venous phases were higher than those of non-recurrent patients, while the ADC value was lower than that of non-recurrent patients. The area under the curves (AUC) of CT value in the arterial phase, CT value in the venous phase, signal intensity value, CNR, and ADC value for the diagnosis of recurrence were 0.828, 0.791, 0.747, 0.801, and 0.801, respectively. The AUC of combined diagnosis of anastomotic recurrence after rectal cancer surgery was 0.920 (95% CI 0.850-0.964), with a sensitivity of 88.71% and a specificity of 87.80%. All of these were superior to the diagnosis of each imaging manifestation and parameter alone (Z=2.485, 2.304, 3.018, 2.313, 3.185, 3.759, 2.706, 2.713, P<0.05), and had good consistency with pathological results (Kappa=0.759, 95% CI 0.566-0.952). Conclusion: Multi-modal imaging combined with CT, HR-MRI and DWI has higher diagnostic value for postoperative anastomotic recurrence of rectal cancer.
Objective: To investigate the value of radiomics based on multiparametric magnetic resonance imaging (MRI) and multiple machine learning algorithms in predicting the expression of Bcl-2 and c-Myc in patients with encephalic primary central nervous system lymphoma (PCNSL). Methods: The clinical data of patients with intracranial PCNSL in Maoming People’s Hospital and Xinyi People’s Hospital from January 2021 to January 2024 were reviewed and analyzed. Based on the expression of Bcl-2 and c-Myc proteins detected by immunohistochemical staining, patients were divided into the double-expression lymphoma (DEL) group and the non-double-expression group (nDEL group). Tumors were manually segmented on MRI images to extract radiomic features. Repeated least absolute shrinkage and selection operator (repeated-LASSO) was applied to select features, followed by the construction of classification models using 15 machine learning algorithms with parameter tuning, custom parameter combinations, LASSO, and 10-fold cross-validation. Results: There were no statistically significant differences in age, gender, presence of hemorrhage or necrosis, tumor location, peritumoral edema, maximum diameter, number of tumors, and presence of meningeal or ependymal invasion between training set, internal validation set, and external validation set. A preliminary set of 2 895 stable radiomic features was obtained based on an intraclass correlation coefficient (ICC>0.75). Repeated-LASSO selected 16 features. The eXtreme Gradient Boosting (XGboost) model and gradient boosting machine (GBM) models, showed the best performance, with the highest area under curve (AUC) of 0.91 in the validation set. Conclusion: Multiparametric MRI combined with multiple machine learning algorithms shows great potential for detecting DEL in PCNSL.
The incidence of thyroid cancer has gradually increased in recent years. The lymph node posterior to the right recurrent laryngeal nerve (LN-prRLN), which is characterized by higher metastasis rate, poor preoperative diagnosis rate, and high correlation with cancer progression, is one of the important pathways for lymph node metastasis in papillary carcinoma of the thyroid (PTC). Due to the complication accompanying LN-prRLN dissection, accurate evaluation of LN-prRLN involvement is significant for the surgical approach and patient prognosis. Ultrasound, as the primary choice for PTC and its cervical lymph node metastasis, plays an important role in the preoperative diagnosis and evaluation of LN-prRLN and has a broad application prospect with the development of artificial intelligence. Conventional ultrasound can directly identify and localize suspicious lymph nodes, combined with contrast-enhanced ultrasound and ultrasound elastography to increase the accuracy of assessment. Lymphatic contrast-enhanced ultrasound is extremely advantageous for the display of LN-prRLN, and emerging ultrasound imaging histology can objectively and quantitatively extract features to construct a joint prediction model for LN-prRLN metastasis to aid in diagnosis. This article reviewed the current status of the application of ultrasound and its new technologies in the preoperative diagnosis and evaluation of LN-prRLN metastasis.
