Training Set Research Articles

Generalized face forgery detection with self-supervised face geometry information analysis network

The emergence of high-quality deepfake facial videos has raised concerns about the security of facial images. Existing face forgery detectors mainly tend to locate a specific forgery region of the human face for detection, which achieves satisfactory performance with known forgery patterns presented in the training set. However, with the continuous advancements in face forgery technology, this approach becomes less reliable with new forgery patterns that emerge. Towards this end, we proposed a novel Self-supervised Face Geometry Information Analysis Network (SF-GAN) method for generalized face forgery detection. SF-GAN effectively leverages the relationships among informative regions based on information theory. Drawing from information theory, regions with high uncertainty tend to contain more valuable information. Our methodology integrates a self-supervised learning mechanism, enabling the precise identification of multiple informative regions. Furthermore, we leverage facial geometry by establishing both explicit and latent geometric relationships through the use of Graph Convolutional Networks (GCNs). Within our framework, facial landmarks and informative regions are depicted as nodes in the GCNs. By analyzing the geometric relationships between the graph of facial landmarks and the graph of informative regions, we are able to identify valid anomalous regions, thereby minimizing uncertainty. Our proposed model gains a comprehensive understanding of common information in face forgery images. Extensive experiments on eight large-scale benchmark datasets: FaceForensics++ (FF++), WildDeepfake (WDF), Celeb-DF v2 (CDF), DeepFake Detection Challenge (DFDC), DFDC preview (DFDC-P), Deepfake Detection (DFD), DeeperForensics-1.0 (DF-1.0) and ForgeryNIR, show that the proposed method is comparable to state-of-the-arts and exhibits better generalizability. Specifically, our SF-GAN, when trained on high-quality FF++ data, achieves an impressive AUC of 76.43% on the CDF dataset.

Individualized prediction of non-sentinel lymph node metastasis in Chinese breast cancer patients with ≥ 3 positive sentinel lymph nodes based on machine-learning algorithms

BackgroundAxillary lymph node dissection (ALND) is a standard procedure for early-stage breast cancer (BC) patients with three or more positive sentinel lymph nodes (SLNs). However, ALND can lead to significant postoperative complications without always providing additional clinical benefits. This study aims to develop machine-learning (ML) models to predict non-sentinel lymph node (non-SLN) metastasis in Chinese BC patients with three or more positive SLNs, potentially allowing the omission of ALND.MethodsData from 2217 BC patients who underwent SLN biopsy at Shantou University Medical College were analyzed, with 634 having positive SLNs. Patients were categorized into those with ≤ 2 positive SLNs and those with ≥ 3 positive SLNs. We applied nine ML algorithms to predict non-SLN metastasis. Model performance was evaluated using ROC curves, precision-recall curves, and calibration curves. Decision Curve Analysis (DCA) assessed the clinical utility of the models.ResultsThe RF model showed superior predictive performance, achieving an AUC of 0.987 in the training set and 0.828 in the validation set. Key predictive features included size of positive SLNs, tumor size, number of SLNs, and ER status. In external validation, the RF model achieved an AUC of 0.870, demonstrating robust predictive capabilities.ConclusionThe developed RF model accurately predicts non-SLN metastasis in BC patients with ≥ 3 positive SLNs, suggesting that ALND might be avoided in selected patients by applying additional axillary radiotherapy. This approach could reduce the incidence of postoperative complications and improve patient quality of life. Further validation in prospective clinical trials is warranted.

Indoor fingerprint localization algorithm based on WKNN and LightGBM-GA

Abstract WiFi-based indoor fingerprint localization is widely used in indoor localization owing to its high accuracy and low deployment costs. Changes in the indoor signal environment directly affect localization accuracy. To improve localization accuracy and stability, this paper proposes a novel indoor fingerprint localization algorithm based on Weighted K-Nearest Neighbors (WKNN) and an enhanced Light Gradient Boosting Machine (LightGBM). First, in the offline phase, Gaussian filtering and K-Nearest Neighbors-Random Forest information completion algorithm with fusion of Euclidean and Manhattan distances are used to remove outliers from the fingerprint database dataset and fill in missing fingerprint information, ensuring the integrity of the fingerprint database. During the online phase, the fingerprint database is divided into training and testing sets. The LightGBM algorithm is used for modeling. Additionally, Genetic Algorithm (GA) is use d to optimize the parameters of LightGBM algorithm to find the best parameters by fitness evaluation. Then, the nearest neighbor set found by the WKNN algorithm is introduced into the LightGBM-GA model. Combining the predictions from the standalone LightGBM algorithm and performing weighted fusion yields the final predicted coordinates. The experiments are conducted in 8 m × 10 m laboratory containing 5 access points and 80 reference points to collect the Received Signal Strength Indication values of 5 WiFi hotspots. The experimental results show that the average localization error of the proposed algorithm is 1.11 m, which is reduced by 6.7%–38.3% compared to K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBoost), LightGBM, KNN + XGBoost, WKNN + LightGBM, and WKNN + XGBoost-GA localization algorithms. The localization curve is smoother, and the cumulative distribution function converges faster. Moreover, the localization time is reduced by 13.3%–36.7%, effectively enhancing localization accuracy and decreasing localization time.

High-quality expert annotations enhance artificial intelligence model accuracy for osteosarcoma X-ray diagnosis.

Primary malignant bone tumors, such as osteosarcoma, significantly affect the pediatric and young adult populations, necessitating early diagnosis for effective treatment. This study developed a high-performance artificial intelligence (AI) model to detect osteosarcoma from X-ray images using highly accurate annotated data to improve diagnostic accuracy at initial consultations. Traditional models trained on unannotated data have shown limited success, with sensitivities of approximately 60%-70%. In contrast, our model used a data-centric approach with annotations from an experienced oncologist, achieving a sensitivity of 95.52%, specificity of 96.21%, and an area under the curve of 0.989. The model was trained using 468 X-ray images from 31 osteosarcoma cases and 378 normal knee images with a strategy to maximize diversity in the training and validation sets. It was evaluated using an independent dataset of 268 osteosarcoma and 554 normal knee images to ensure generalizability. By applying the U-net architecture and advanced image processing techniques such as renormalization and affine transformations, our AI model outperforms existing models, reducing missed diagnoses and enhancing patient outcomes by facilitating earlier treatment. This study highlights the importance of high-quality training data and advocates a shift towards data-centric AI development in medical imaging. These insights can be extended to other rare cancers and diseases, underscoring the potential of AI in transforming diagnostic processes in oncology. The integration of this AI model into clinical workflows could support physicians in early osteosarcoma detection, thereby improving diagnostic accuracy and patient care.

Enhancing protein-ligand binding affinity prediction through sequential fusion of graph and convolutional neural networks.

Predicting protein-ligand binding affinity is a crucial and challenging task in structure-based drug discovery. With the accumulation of complex structures and binding affinity data, various machine-learning scoring functions, particularly those based on deep learning, have been developed for this task, exhibiting superiority over their traditional counterparts. A fusion model sequentially connecting a graph neural network (GNN) and a convolutional neural network (CNN) to predict protein-ligand binding affinity is proposed in this work. In this model, the intermediate outputs of the GNN layers, as supplementary descriptors of atomic chemical environments at different levels, are concatenated with the input features of CNN. The model demonstrates a noticeable improvement in performance on CASF-2016 benchmark compared to its constituent CNN models. The generalization ability of the model is evaluated by setting a series of thresholds for ligand extended-connectivity fingerprint similarity or protein sequence similarity between the training and test sets. Masking experiment reveals that model can capture key interaction regions. Furthermore, the fusion model is applied to a virtual screening task for a novel target, PI5P4Kα. The fusion strategy significantly improves the ability of the constituent CNN model to identify active compounds. This work offers a novel approach to enhancing the accuracy of deep learning models in predicting binding affinity through fusion strategies.

CRML-Net: Cross-Modal Reasoning and Multi-Task Learning Network for tooth image segmentation

Data from a single modality may suffer from noise, low contrast, or other imaging limitations that affect the model’s accuracy. Furthermore, due to the limited amount of data, most models trained on single-modality data tend to overfit the training set and perform poorly on out-of-domain data. Therefore, in this paper, we propose a network named Cross-Modal Reasoning and Multi-Task Learning Network (CRML-Net), which combines cross-modal reasoning and multi-task learning, aiming to leverage the complementary information between different modalities and tasks to enhance the model’s generalization ability and accuracy. Specifically, CRML-Net consists of two stages. In the first stage, our network extracts a new morphological information modality from the original image and then performs cross-modal fusion with the original modality image, aiming to leverage the morphological information to enhance the model’s robustness to out-of-domain datasets. In the second stage, based on the output of the previous stage, we introduce a multi-task learning mechanism, aiming to improve the model’s performance on unseen data by sharing surface detail information from auxiliary tasks. We validated our method on a publicly available tooth cone beam computed tomography dataset. Our evaluation demonstrates that our method outperforms state-of-the-art approaches.

Development and validation of a nomogram for predicting rapid relapse in triple-negative breast cancer patients treated with neoadjuvant chemotherapy.

Triple-negative breast cancer (TNBC) accounts for disproportionately poor outcomes in breast cancer, driven by a subset of rapid-relapse TNBC (rrTNBC) with marked chemoresistance, rapid metastatic spread, and poor survival. This study aimed to develop and validate a nomogram based on clinicopathological characteristics to predict rapid relapse in TNBC patients treated with neoadjuvant chemotherapy (NAC) first. The clinicopathological data of 504 TNBC patients treated with NAC first in Tianjin Medical University Cancer Hospital were analyzed retrospectively, with 109 rapid relapsed patients, and 395 non-rapid relapsed patients, respectively. Based on clinicopathologic characteristics, and follow-up data were analyzed. The independent predictors of clinicopathological characteristics were identified by logistic regression analysis and then used to build a nomogram. The concordance index (C-index), the area under the curve (AUC) of receiver operating characteristic (ROC), and calibration plots were used to evaluate the performance of the model. Univariate and multivariate logistic regression analyses showed that age at diagnosis (age≥50years, OR = 0.325,95% CI:0.137-0.771), Nodal staging (N3 staging, OR = 13.669,95% CI:3.693-50.592),sTIL expression levels (sTIL intermediate expression, OR = 0.272,95% CI:0.109-0.678; sTIL high expression, OR = 0.169,95% CI:0.048-0.594), and NAC response (ORR, OR = 0.059,95% CI:0.024-0.143) were independent predictors of rapid relapse in TNBC patients treated with NAC firstly. Among these independent predictors, age ≥ 50years, sTIL intermediate expression, sTIL high expression, and ORR in NAC were independent protective factors for rapid relapse in TNBC NAC patients. N3 staging was an independent risk factor for rapid relapse in TNBC NAC patients. The ROC curve, calibration curve, and decision curve analysis were used to validate the model. The C-Index of the training sets and validation sets were 0.938 and 0.910, respectively. The Brier scores of the training sets and validation sets were 0.076 and 0.097, respectively. This study developed and verified a nomogram for predicting rapid relapse in TNBC NAC patients, and the predictive model had high discrimination and accuracy.

Risk assessment tool for anemia of chronic disease in systemic lupus erythematosus: a prediction model.

This study aims to develop a predictive model for estimating the likelihood of anemia of chronic disease (ACD) in patients with systemic lupus erythematosus (SLE) and to elucidate the relationship between various factors and ACD METHODS: Individuals diagnosed with SLE for at least one year were enrolled and categorized into two groups: those with ACD and those without anemia symptoms. Patients were randomly assigned to training and test sets at an 8:2 ratio. The least absolute shrinkage and selection operator (LASSO) method was used to select predictors, followed by logistic regression for modeling. Model performance was assessed using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA) for both training and test sets. The study included a total of 216 patients, with 172 in the training set and 44 in the test set. LASSO identified 6 variables for constructing the predictive model, resulting in an area under the curve (AUC) of 0.833 (95% CI, 0.773-0.892) in the training set and 0.861 (95% CI, 0.750-0.972) in the test set. Calibration curves indicated consistency between expected and observed probabilities. DCA indicated that the model yielded a net benefit with threshold probabilities ranging from 20% to 90% in the training set and from 10% to 80% in the test set. This study presents a predictive model for assessing the risk of ACD in SLE patients. The model effectively captures the underlying mechanism of ACD in SLE and empowers clinicians to make well-informed treatment adjustments. Key Points • Development of a New Predictive Model: This study introduces a new predictive model to evaluate the likelihood of anemia of chronic disease (ACD) in patients with systemic lupus erythematosus (SLE). The model utilizes routine laboratory parameters to identify high-risk individuals, addressing a significant gap in current clinical practice. • Reflection of Potential Mechanisms for ACD Development: By incorporating the factors needed to construct the predictive model, this study also sheds light on the potential mechanisms of ACD development in SLE patients.

Development and Validation of a Contrast-Enhanced US VI-RADS for Evaluating Muscle Invasion in Bladder Cancer.

Background Contrast-enhanced US (CEUS) can be used preoperatively for evaluating muscle invasion in bladder cancer, which is important for determining appropriate treatment. However, diagnostic criteria for assessing this at CEUS have not been standardized. Purpose To develop and validate a CEUS Vesical Imaging Reporting and Data System (VI-RADS) for evaluating muscle invasion in bladder cancer. Materials and Methods This single-center prospective study consecutively enrolled patients with suspected bladder cancer. Participants underwent transabdominal or intracavity CEUS between July 2021 and May 2023. Participants were divided into a training set and a validation set at a 2:1 ratio based on the chronologic order of enrollment. The training set was used to identify major imaging features to include in CEUS VI-RADS, and the likelihood of muscle invasion per category was determined using a pathologic reference standard. The optimal VI-RADS category cutoff for muscle invasion was determined with use of the maximum Youden index. The validation set was assessed by novice and expert readers and used to validate the diagnostic performance and interreader agreement of the developed system. Results Overall, 126 participants (median age, 64 years [IQR, 57-71 years]; 107 male) and 67 participants (median age, 64 years [IQR, 56-69 years]; 49 male) were included in the training and validation set, respectively. In the training set, the optimal CEUS VI-RADS category cutoff for muscle invasion was VI-RADS 4 or higher (Youden index, 0.77). In the validation set, CEUS VI-RADS achieved good performance for both novice and expert readers (area under the receiver operating characteristic curve, 0.80 [95% CI: 0.70, 0.90] vs 0.88 [95% CI: 0.80, 0.97]; P = .09). The interreader agreement regarding the evaluation of CEUS VI-RADS category was 0.77 (95% CI: 0.65, 0.85) for novice readers, 0.87 (95% CI: 0.79, 0.92) for expert readers, and 0.78 (95% CI: 0.70, 0.84) for all readers. Conclusion The developed CEUS VI-RADS showed good performance and interreader agreement for the assessment of muscle invasion in bladder cancer. Chinese Clinical Trial Registry no. ChiCTR2100049435 © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Morrell in this issue.

Preoperative prediction model of lymph node metastasis in the inguinal and femoral region based on radiomics and artificial intelligence

ObjectiveTo predict preoperative inguinal lymph node metastasis in vulvar cancer patients using a machine learning model based on imaging features and clinical data from pelvic magnetic resonance imaging (MRI).Methods52 vulvar...

Development and Validation of a Nomogram for Predicting Breast Malignancy in Male Patients Based on Clinical and Ultrasound Features.

This study aimed to construct a nomogram based on clinical and ultrasound (US) features to predict breast malignancy in males. The medical records between August, 2021 and February, 2023 were retrospectively collected from the database. Patients included in this study were randomly divided into training and validation sets in a 7:3 ratio. The models for predicting the risk of malignancy in male patients with breast lesions were virtualized by the nomograms. Among the 71 enrolled patients, 50 were grouped into the training set, while 21 were grouped into the validation set. After the multivariate analysis was done, pain, BI-RADS category, and elastography score were identified as the predictors for malignancy risk and were selected to generate the nomogram. The C-index was 0.931 for the model. Concordance between predictions and observations was detected by calibration curves and was found to be good in this study. The model achieved a net benefit across all threshold probabilities, which was shown by the decision curve analysis (DCA) curve. We successfully constructed a nomogram to evaluate the risk of breast malignancy in males using clinical and US features, including pain, BI-RADS category, and elastography score, which yielded good predictive performance.

Prediction of high-risk prostate cancer based on the habitat features of biparametric magnetic resonance and the omics features of contrast-enhanced ultrasound

Rationale and objectivesTo predict high-risk prostate cancer (PCa) by combining the habitat features of biparametric magnetic resonance imaging (bp-MRI) with the omics features of contrast-enhanced ultrasound (CEUS). Materials and methodsThis study retrospectively collected patients with PCa confirmed by histopathology from January 2020 to June 2023. All patients underwent bp-MRI and CEUS of the prostate, followed by a targeted and transrectal systematic prostate biopsy. The cases were divided into the intermediate-low-risk group (Gleason score ≤7, n = 59) and high-risk group (Gleason score ≥8, n = 33). Radiomics prediction models, namely, MRI_habitat, CEUS_intra, and MRI-CEUS models, were developed based on the habitat features of bp-MRI, the omics features of CEUS, and a merge of features of the two, respectively. Predicted probabilities, called radscores, were then obtained. Clinical-radiological indicators were screened to construct clinic models, which generated clinic scores. The omics–clinic model was constructed by combining the radscore of MRI-CEUS and the clinic score. The predictive performance of all the models was evaluated using the receiver operating characteristic curve. ResultsThe area under the curve (AUC) values of the MRI-CEUS model were 0.875 and 0.842 in the training set and test set, respectively, which were higher than those of the MR_habitat (training set: 0.846, test set: 0.813), CEUS_intra (training set: 0.801, test set: 0.743), and clinic models (training set: 0.722, test set: 0.611). The omics–clinic model achieved a higher AUC (train set: 0.986, test set: 0.898). ConclusionsThe combination of the habitat features of bp-MRI and the omics features of CEUS can help predict high-risk PCa.

A Comparative Study on the Diagnostic Performance of Multiple Radiomics Models in Differentiating PGL and MIA in Pulmonary Ground-Glass Nodules

Rationale and ObjectivesHistological subtypes of lung cancers are critical for clinical treatment decision. The aim of this study is to compare the diagnostic performance of multiple radiomics models in differentiating PGL and MIA in pulmonary GGN, in order to identify the most optimal diagnostic model. Materials and methodsPatients presenting with GGNs on lung CT, confirmed as PGL or MIA through surgical pathology between October 2015 and June 2023, were included. The GGNs were randomly divided into training and validation sets at a 7:3 ratio. Clinical imaging characteristics were analyzed by univariate and multivariate logistic regression to identify independent risk factors for predicting MIA, leading to the development of a clinical model. ITK-SNAP and Pyradiomics were employed for segmentation and radiomics feature extraction. Subsequently, radiomics and combined models were established. The diagnostic performance of the three models was compared using ROC curves and quantitatively assessed by AUC, accuracy, specificity, and sensitivity. ResultsA total of 116 cases of GGNs with pathologically confirmed PGLs and MIAs were included. The clinical model identified three independent predictors. The radiomics model identified seven distinct radiomic features. A combined model was constructed by integrating clinical imaging features with radiomic features. In the training set, the combined model demonstrated a higher AUC than the radiomics model, with AUCs of 0.87 and 0.85 respectively. In the validation set, the radiomics model outperformed the combined model with an AUC of 0.83 versus 0.82. Notably, the radiomics model achieved the highest accuracy and specificity, while the combined model demonstrated the highest sensitivity. However, both models performed significantly better than the clinical model. ConclusionThe independent radiomics model can serve as a rapid, non-invasive diagnostic tool for differentiating between the PGL and MIA.

Standardized IETA criteria enhance accuracy of junior and intermediate ultrasound radiologists in diagnosing malignant endometrial and intrauterine lesions.

To transform the standardized descriptions of the ultrasound characteristics of endometrial and intrauterine lesions devised by the International Endometrial Tumor Analysis (IETA) group into a practical scoring method and to investigate whether application of this method enhances the diagnostic accuracy of ultrasound radiologists with different levels of experience in detecting malignancy compared with subjective assessment. This was a retrospective study of 855 patients with endometrial and/or intrauterine lesions, who were divided into a training (n = 600) and a validation (n = 255) set. Ultrasound radiologists with varying levels of experience (expert, intermediate and junior) evaluated all lesions by subjective assessment and according to IETA rules. Using IETA rules, the experts identified signs of malignancy in the training set, assigned scores for each indicator and validated the scoring method in the validation set. The intermediate-level and junior ultrasound radiologists reassessed the malignancy of the lesions using the IETA scoring method and compared their classifications with those made previously by subjective assessment. Postsurgical pathological evaluation was used as the reference standard. Using subjective assessment, the experts demonstrated the highest level of diagnostic accuracy, with a sensitivity of 85.0%, specificity of 94.3% and an area under the receiver-operating-characteristics curve (AUC) of 0.897. Applying the IETA scoring method (comprising eight ultrasound characteristics that contributed to the total score) with a threshold of > 25 points for the diagnosis of malignancy achieved a sensitivity of 84.7%, specificity of 94.7% and AUC of 0.9533 in the training set, with similar performance in the validation set, when performed by experts. Using the IETA scoring method, both junior and intermediate ultrasound radiologists showed improvement in sensitivity (from 55.5% to 74.8% and from 70.2% to 77.1%, respectively), specificity (from 88.4% to 91.5% and from 87.4% to 92.2%, respectively) and AUC (from 0.704 to 0.827 and from 0.793 to 0.841, respectively) for diagnosing malignant lesions. The IETA scoring method exhibits high diagnostic efficacy for malignant endometrial and intrauterine lesions. This method compensates for the lack of experience among junior and intermediate-level ultrasound radiologists, enhancing their diagnostic skill to a level nearing that of experienced senior ultrasound radiologists. Further research is essential to validate the practicality of implementing this method and to confirm its clinical value. © 2024 International Society of Ultrasound in Obstetrics and Gynecology.

Establishment and validation of a nomogram model containing a triglyceride-glucose index and neutrophil-to-high-density lipoprotein ratio for predicting major adverse cardiac events in patients with ST-segment elevation myocardial infarction.

To analyze the predictive value of the triglyceride-glucose (TyG) index and neutrophil-to-high-density lipoprotein ratio (NHR) for in-hospital major adverse cardiac events (MACE) after percutaneous coronary intervention (PCI) in patients with acute ST-segment elevation myocardial infarction (STEMI), and to establish an associated nomogram model. In this retrospective study, we collected data from consecutive STEMI patients who underwent PCI from October 2019 to June 2023 at the Second People's Hospital of Hefei and the Second Affiliated Hospital of Anhui Medical University, as training and validation sets. Stepwise regression and multivariate logistic regression analysis were performed to screen independent risk factors, and a nomogram model was constructed and evaluated for its predictive efficacy. The TyG index, NHR, urea, diastolic blood pressure, hypertension, and left ventricular ejection fraction were independent risk factors for in-hospital MACE after PCI, and were used to construct the nomogram model. The C-index of the training and validation sets were 0.799 and 0.753, respectively, suggesting that the model discriminated well. Calibration and clinical decision curves also demonstrated that the nomogram model had good predictive power. In STEMI patients, increased TyG index and NHR were closely related to the occurrence of in-hospital MACE after PCI. Our constructed nomogram model has some value for predicting the occurrence of in-hospital MACE in STEMI patients.

A Novel Ultrasound-Based Radiomics Model for the Preoperative Prediction of Lymph Node Metastasis in Cervical Cancer

ObjectiveThe purpose of this retrospective study was to establish a combined model based on ultrasound (US)-radiomics and clinical factors to predict preoperative lymph node metastasis (LNM) in cervical cancer (CC) patients non-invasively. MethodsA total of 131 CC patients who had cervical lesions found by transvaginal sonography (TVS) from the First Affiliated Hospital of Anhui Medical University (Hefei, China) were retrospectively analyzed. The clinical independent predictors were selected using univariate and multivariate logistic regression analysis. US-radiomics features were extracted from US images; after selecting the most significant features by univariate analysis, Spearman's correlation analysis, and the least absolute shrinkage and selection operator (LASSO) algorithm; four machine-learning classification algorithms were used to build the US-radiomics model. Fivefold cross-validation was then used to test the performance of the model and compare the ability of the clinical, US-radiomics and combined models to predict LNM in CC patients. ResultsRed blood cell, platelet and squamous cell carcinoma-associated antigen were independent clinical predictors of LNM (+) in CC patients. eXtreme Gradient Boosting performed the best among the four machine-learning classification algorithms. Fivefold cross-validation confirmed that eXtreme Gradient Boosting indeed performs the best, with average area under the curve values in the training and validation sets of 0.897 and 0.898. In the three prediction models, both the US-radiomics model and the combined model showed good predictive efficacy, with average area under the curve values in the training and validation sets of 0.897, 0.898 and 0.912, 0.905, respectively. ConclusionUS-radiomics features combined with clinical factors can preoperatively predict LNM in CC patients non-invasively.

Development and Validation of Upper Limb Lymphedema in Patients After Breast Cancer Surgery Using a Practicable Machine Learning Model: A Retrospective Cohort Study.

Upper limb lymphedema is one of the most common adverse events related to surgery owing to the large gap between guideline implementation and the intended clinical outcomes. However, the monitoring of limb lymphedema remains challenging because of vague clinical presentations. This study aimed to develop and validate practical predictive models for upper limb lymphedema through machine learning. We retrospectively collected clinical data to develop models for early risk prediction of upper limb lymphedema based on a single-center electronic health record data from patients who underwent breast cancer surgery from June 2021 through June 2023. For prediction model building, 70% and 30% of the data were randomly split into training and testing sets, respectively. We then developed an upper limb lymphedema prediction model using machine learning algorithms, which included random forest model (RFM), generalized logistic regression model (GLRM), and artificial neural network model (ANNM). For evaluating the model's performance, we used the area under the receiver operating characteristic curve (AUROC), calibration curve to compare different models. The potential clinical usefulness of the best model at the best threshold was assessed through a net benefit approach using a decision curve analysis (DCA). Of the 3201 patients screened for eligibility, 3160 participants were recruited for the prediction model. Among these, Body Mass Index (BMI), hypertension, TNM, lesion site, level of lymph node dissection(LNMD), treatment, and nurse were independent risk factors for upper limb lymphedema and were listed as candidate variables of ML-based prediction models. The RFM algorithm, in combination with seven candidate variables, demonstrated the highest prediction efficiency in both the training and internal verification sets, with an area under the curve (AUC) of 0.894 and 0.889 and a 95% confidence interval (CI) of 0.839-0.949 and 0.834-0.944, respectively. The other two types of prediction models had prediction efficiencies between AUCs of 0.731 and 0.819 and 95% CIs of 0.674-0.789 and 0.762-0.876, respectively. The interpretable predictive model helps physicians more accurately predict the upper limb lymphedema risk in patients undergoing breast cancer surgery. Especially for the RFM, this newly established machine learning-based model has shown good predictive ability for distinguishing high risk of upper limb lymphedema, which could facilitate future clinical decisions, hospital management, and improve outcomes.

Predicting costs of local public bus transport services through machine learning methods

The present study developed several machine learning-based cost models to predict an efficient total economic cost per vehicle revenue-mile of urban public bus transport. The models were trained on a built-in dataset from 269 transit agencies providing urban services in the United States from 2015 to 2019. A feature selection strategy was implemented, finding that, for each proposed model, a subset of features determined a large impact on unit cost. These “core” features included commercial speed, average salary expenses per employee, vehicle productivity, and fleet ownership cost per vehicle. Machine learning techniques outperformed the linear regression method in terms of predictive power and robustness (understood as the dispersion of predictive power measures over the training sets). Based on SHAP values, the sensitivity analyses showed that the proposed models could be used to predict the impact of changes in some critical features on corresponding unit costs. The results may be useful for: (i) introducing regulatory constraints to the allocation of national public resources to local public bus transport services, aimed at minimizing the resources needed to provide a given level of service; (ii) defining the maximum economic compensation required by firms involved in competitive tendering for the allotment of service concessions, or firms with monopoly rights (by political choice and/or local public ownership); and (iii) improving service contract management and design by identifying key cost drivers of transit services.

Identification of Molecular Correlations of GSDMD with Pyroptosis inAlzheimer's Disease.

An analysis of bioinformatics and cell experiments was performed to verify the relationship between gasdermin D (GSDMD), an executive protein of pyroptosis, and Alzheimer's disease (AD). The training set GSE33000 was utilized to identify differentially expressed genes (DEGs) in both the AD group and control group, as well as in the GSDMD protein high/low expression group. Subsequently, the weighted gene co-expression network analysis (WGCNA) and the least absolute shrinkage and selection operator (LASSO) regression analysis were conducted, followed by the selection of the key genes for the subsequent Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses. The association between GSDMD and AD was assessed and confirmed in the training set GSE33000, as well as in the validation sets GSE5281 and GSE48350. Immunofluorescence (IF) was employed to detect the myelin basic protein (MBP), a distinctive protein found in the rat oligodendrocytes (OLN-93 cells). A range of concentrations (1-15 μmol/L) of β-amyloid 1-42 (Aβ1-42) were exposed to the cells, and the subsequent observations were made regarding cell morphology. Additionally, the assessments were conducted to evaluate the cell viability, the lactate dehydrogenase (LDH) release, the cell membrane permeability, and the GSDMD protein expression. A total of 7,492 DEGs were screened using GSE33000. Subsequently, WGCNA analysis identified 19 genes that exhibited the strongest correlation with clinical traits in AD. Additionally, LASSO regression analysis identified 13 key genes, including GSDMD, AFF1, and ATOH8. Furthermore, the investigation revealed that the key genes were associated with cellular inflammation based on GO and KEGG analyses. Moreover, the area under the curve (AUC) values for the key genes in the training and validation sets were determined to be 0.95 and 0.70, respectively. Significantly, GSDMD demonstrated elevated levels of expression in AD across both datasets. The positivity of MBP expression in cells exceeded 95%. As the concentration of Aβ1-42 action gradually escalated, the detrimental effects on cells progressively intensified, resulting in a gradual decline in cell survival rate, accompanied by an increase in lactate dehydrogenase release, cell membrane permeability, and GSDMD protein expression. The association between GSDMD and AD has been observed, and it has been found that Aβ1-42 can induce a significant upregulation of GSDMD in OLN-93 cells. This suggests that Aβ1-42 has the potential to induce cellular pyroptosis and can serve as a valuable cellular pyroptosis model for the study of AD.

Image-driven prediction of fatigue crack growth in metal materials via spatiotemporal neural network

This study proposes an image-driven model based on the SimVP spatiotemporal neural network (STNN) to predict the fatigue crack growth (FCG) in aluminum alloys. This methodology represents a novel usage of STNNs for FCG analysis. It does not require repetitive modeling, extensive computations, or conventional mechanical assumptions. The datasets used during this study were gathered from fatigue experiments with a variety of crack positions, angles, and load levels; they contained a total of 17,925 image frames obtained from DIC measurements. Subsequently, the displacement fields were interpolated onto uniform grids and then augmented, so they could be fitted into an STNN. The proposed method was validated using specimens with edge and central cracks subjected to loads equal to 15.0 % and 20.0 % of the ultimate load. The generalization capability of the proposed method was studied by predicting the FCG under load levels and crack angles outside the training set. In addition, its predictive capability was investigated for both short and long step sizes by employing datasets in which the image data were collected at varying intervals. The overall structural similarity index measurement values were greater than 0.968, and the root mean square errors were held within 0.025 mm. The predicted displacement fields, crack lengths, and crack growth rates agreed well with experimental measurements.