Deep learning-assisted diagnosis of large vessel occlusion in acute ischemic stroke based on four-dimensional computed tomography angiography.

Purpose: To develop deep learning models based on four-dimensional computed tomography angiography (4D-CTA) images for automatic detection of large vessel occlusion (LVO) in the anterior circulation that cause acute ischemic stroke. Methods: This retrospective study included 104 LVO patients and 105 non-LVO patients for deep learning models development. Another 30 LVO patients and 31 non-LVO patients formed the time-independent validation set. Four phases of 4D-CTA (arterial phase P1, arterial-venous phase P2, venous phase P3 and late venous phase P4) were arranged and combined and two input methods was used: combined input and superimposed input. Totally 26 models were constructed using a modified HRNet network. Assessment metrics included the areas under the curve (AUC), accuracy, sensitivity, specificity and F1 score. Kappa analysis was performed to assess inter-rater agreement between the best model and radiologists of different seniority. Results: The P1 + P2 model (combined input) had the best diagnostic performance. In the internal validation set, the AUC was 0.975 (95%CI: 0.878-0.999), accuracy was 0.911, sensitivity was 0.889, specificity was 0.944, and the F1 score was 0.909. In the time-independent validation set, the model demonstrated consistently high performance with an AUC of 0.942 (95%CI: 0.851-0.986), accuracy of 0.902, sensitivity of 0.867, specificity of 0.935, and an F1 score of 0.901. The best model showed strong consistency with the diagnostic efficacy of three radiologists of different seniority (k = 0.84, 0.80, 0.70, respectively). Conclusion: The deep learning model, using combined arterial and arterial-venous phase, was highly effective in detecting LVO, alerting radiologists to speed up the diagnosis.

Quantitative Imaging for Predicting Hematoma Expansion in Intracerebral Hemorrhage: A Multimodel Comparison.

BACKGROUND: Several studies report that radiomics provides additional information for predicting hematoma expansion in intracerebral hemorrhage (ICH). However, the comparison of diagnostic performance of radiomics for predicting revised hematoma expansion (RHE) remains unclear. METHODS: The cohort comprised 312 consecutive patients with ICH. A total of 1106 radiomics features from seven categories were extracted using Python software. Support vector machines achieved the best performance in both the training and validation datasets. Clinical factors models were constructed to predict RHE. Receiver operating characteristic curve analysis was used to assess the abilities of non-contrast computed tomography (NCCT) signs, radiomics features, and combined models to predict RHE. RESULTS: We finally selected the top 21 features for predicting RHE. After univariate analysis, 4 clinical factors and 5 NCCT signs were selected for inclusion in the prediction models. In the training and validation dataset, radiomics features had a higher predictive value for RHE (AUC = 0.83) than a single NCCT sign and expansion-prone hematoma. The combined prediction model including radiomics features, clinical factors, and NCCT signs achieved higher predictive performances for RHE (AUC = 0.88) than other combined models. CONCLUSIONS: NCCT radiomics features have a good degree of discrimination for predicting RHE in ICH patients. Combined prediction models that include quantitative imaging significantly improve the prediction of RHE, which may assist in the risk stratification of ICH patients for anti-expansion treatments.

ETUNet:Exploring efficient transformer enhanced UNet for 3D brain tumor segmentation.

Medical image segmentation is a crucial topic in medical image processing. Accurately segmenting brain tumor regions from multimodal MRI scans is essential for clinical diagnosis and survival prediction. However, similar intensity distributions, variable tumor shapes, and fuzzy boundaries pose severe challenges for brain tumor segmentation. Traditional segmentation networks based on UNet struggle to establish explicit long-range dependencies from the feature space due to the limitations of the CNN receptive field. This is particularly crucial for dense prediction tasks such as brain tumor segmentation. Recent works have incorporated the powerful global modeling capability of Transformer into UNet to achieve more precise segmentation results. Nevertheless, these methods encounter some issues: (1) the global information is often modeled by simply stacking Transformer layers for a specific module, resulting in high computational complexity and underutilization of the potential of the UNet architecture; (2) the rich boundary information of tumor subregions in multi-scale features is often overlooked. Motivated by these challenges, we propose an advanced fusion of Transformer with UNet by reexamining the core three parts (encoder, bottleneck, and skip connections). Firstly, we introduce a CNN-Transformer module in the encoder to replace the traditional CNN module, enabling the capture of deep spatial dependencies from input images. To address high-level semantic information, we incorporate a computationally efficient spatial-channel attention layer in the bottleneck for global interaction, highlighting important semantic features from the encoder path output. For irregular lesions, we fuse the multi-scale features from the encoder output and the decoder features in the skip connections by calculating cross-attention. This adaptive querying of valuable information from multi-scale features enhances the boundary localization ability of the decoder path and suppresses redundant features with low correlation. Compared to existing methods, our model further enhances the learning capacity of the overall UNet architecture while maintaining low computational complexity. Experimental results on the BraTS2018 and BraTS2020 datasets for brain tumor segmentation tasks demonstrate that our model achieves comparable or superior results compared to recent CNN or Transformer-based models. The average DSC and HD95 on the two datasets are 0.854, 6.688, and 0.862, 5.455 respectively. At the same time, our model achieves optimal segmentation of Enhancing tumors, showcasing the effectiveness of our method. Our code will be made publicly available at https://github.com/wzhangck/ETUnet.

Radiomics derived from T2-FLAIR: the value of 2- and 3-classification tasks for different lesions in multiple sclerosis.

Background: White matter (WM) lesions can be classified into contrast enhancement lesions (CELs), iron rim lesions (IRLs), and non-iron rim lesions (NIRLs) based on different pathological mechanism in relapsing-remitting multiple sclerosis (RRMS). The application of radiomics established by T2-FLAIR to classify WM lesions in RRMS is limited, especially for 3-class classification among CELs, IRLs, and NIRLs. Methods: A total of 875 WM lesions (92 CELs, 367 IRLs, 416 NIRLs) were included in this study. The 2-class classification was only performed between IRLs and NIRLs. For the 2- and 3-class classification tasks, all the lesions were randomly divided into training and testing sets with a ratio of 8:2. We used least absolute shrinkage and selection operator (LASSO), reliefF algorithm, and mutual information (MI) for feature selection, then eXtreme gradient boosting (XGBoost), random forest (RF), and support vector machine (SVM) were used to establish discrimination models. Finally, the area under the curve (AUC), accuracy, sensitivity, specificity, and precision were used to evaluate the performance of the models. Results: For the 2-class classification model, LASSO classifier with RF model showed the best discrimination performance with the AUC of 0.893 (95% CI: 0.838-0.942), accuracy of 0.813, sensitivity of 0.833, specificity of 0.781, and precision of 0.851. However, the 3-class classification model of LASSO with XGBoost displayed the highest performance with the AUC of 0.920 (95% CI: 0.887-0.950), accuracy of 0.796, sensitivity of 0.839, specificity of 0.881, and precision of 0.846. Conclusions: Radiomics models based on T2-FLAIR images have the potential for discriminating among CELs, IRLs, and NIRLs in RRMS.

LVONet: automatic classification model for large vessel occlusion based on the difference information between left and right hemispheres.

Objective.Stroke is a highly lethal condition, with intracranial vessel occlusion being one of its primary causes. Intracranial vessel occlusion can typically be categorized into four types, each requiring different intervention measures. Therefore, the automatic and accurate classification of intracranial vessel occlusions holds significant clinical importance for assessing vessel occlusion conditions. However, due to the visual similarities in shape and size among different vessels and variations in the degree of vessel occlusion, the automated classification of intracranial vessel occlusions remains a challenging task. Our study proposes an automatic classification model for large vessel occlusion (LVO) based on the difference information between the left and right hemispheres.Approach.Our approach is as follows. We first introduce a dual-branch attention module to learn long-range dependencies through spatial and channel attention, guiding the model to focus on vessel-specific features. Subsequently, based on the symmetry of vessel distribution, we design a differential information classification module to dynamically learn and fuse the differential information of vessel features between the two hemispheres, enhancing the sensitivity of the classification model to occluded vessels. To optimize the feature differential information among similar vessels, we further propose a novel cooperative learning loss function to minimize changes within classes and similarities between classes.Main results.We evaluate our proposed model on an intracranial LVO data set. Compared to state-of-the-art deep learning models, our model performs optimally, achieving a classification sensitivity of 93.73%, precision of 83.33%, accuracy of 89.91% and Macro-F1 score of 87.13%.Significance.This method can adaptively focus on occluded vessel regions and effectively train in scenarios with high inter-class similarity and intra-class variability, thereby improving the performance of LVO classification.

A fusion model integrating magnetic resonance imaging radiomics and deep learning features for predicting alpha-thalassemia X-linked intellectual disability mutation status in isocitrate dehydrogenase-mutant high-grade astrocytoma: a multicenter study.

Background: The mutational status of alpha-thalassemia X-linked intellectual disability (ATRX) is an important indicator for the treatment and prognosis of high-grade gliomas, but reliable ATRX testing currently requires invasive procedures. The objective of this study was to develop a clinical trait-imaging fusion model that combines preoperative magnetic resonance imaging (MRI) radiomics and deep learning (DL) features with clinical variables to predict ATRX status in isocitrate dehydrogenase (IDH)-mutant high-grade astrocytoma. Methods: A total of 234 patients with IDH-mutant high-grade astrocytoma (120 ATRX mutant type, 114 ATRX wild type) from 3 centers were retrospectively analyzed. Radiomics and DL features from different regions (edema, tumor, and the overall lesion) were extracted to construct multiple imaging models by combining different features in different regions for predicting ATRX status. An optimal imaging model was then selected, and its features and linear coefficients were used to calculate an imaging score. Finally, a fusion model was developed by combining the imaging score and clinical variables. The performance and application value of the fusion model were evaluated through the comparison of receiver operating characteristic curves, the construction of a nomogram, calibration curves, decision curves, and clinical application curves. Results: The overall hybrid model constructed with radiomics and DL features from the overall lesion was identified as the optimal imaging model. The fusion model showed the best prediction performance with an area under curve of 0.969 in the training set, 0.956 in the validation set, and 0.949 in the test set as compared to the optimal imaging model (0.966, 0.916, and 0.936, respectively) and clinical model (0.677, 0.641, 0.772, respectively). Conclusions: The clinical trait-imaging fusion model based on preoperative MRI could effectively predict the ATRX mutation status of individuals with IDH-mutant high-grade astrocytoma and has the potential to help patients through the development of a more effective treatment strategy before treatment.

Radiomics models based on cortical damages for identification of multiple sclerosis with cognitive impairment.

BACKGROUND: Cognitive impairment (CI) is a common symptom in multiple sclerosis (MS) patients. Cortical damages can be closely associated with cognitive network dysfunction and clinically significant CI in MS. So, in this study, We aimed to develop a radiomics model to efficiently identify the MS patients with CI based on clinical data and cortical damages. METHODS: One hundred and eighteen patients with MS were divided into CI and normal cognitive (NC) cohorts (62/56) as defined by the Montreal Cognitive Assessment (MoCA). All participants were randomly divided into train and test sets with a ratio of 7:3. The radiomic features were selected by using the least absolute shrinkage and selection operator (LASSO) method. The discrimination models were built with the support vector machines (SVM) by the clinical data, radiomic features, and merge data, respectively. And the patients were further divided according to each cognitive domain including memory, visuospatial, language, attention and executive, and each domain model was applied by the most suitable classifier. RESULTS: A total of 2298 features were extracted, of which 36 were finally selected. The merge model showed the greatest performance with the area under the curve (AUC) of 0.86 (95 % confidence interval: 0.81-0.91), accuracy (ACC) of 0.78, sensitivity of 0.79 and specificity of 0.77 in test cohort. However, although the visuospatial domain model showed the highest AUC of 0.71 (95 % confidence interval: 0.61-0.81) among five domain models, other domain models did not meet satisfactory results with a relatively low AUC, ACC, sensitivity and specificity. CONCLUSIONS: The radiomics model based on clinical data and cortical damages had a great potential to identify the MS patients with CI for clinical cognitive assessment.

Preoperative Discrimination of CDKN2A/B Homozygous Deletion Status in Isocitrate Dehydrogenase-Mutant Astrocytoma: A Deep Learning-Based Radiomics Model Using MRI.

BACKGROUND: Cyclin-dependent kinase inhibitor 2A/B (CDKN2A/B) homozygous deletion has been verified as an independent and critical biomarker of negative prognosis and short survival in isocitrate dehydrogenase (IDH)-mutant astrocytoma. Therefore, noninvasive and accurate discrimination of CDKN2A/B homozygous deletion status is essential for the clinical management of IDH-mutant astrocytoma patients. PURPOSE: To develop a noninvasive, robust preoperative model based on MR image features for discriminating CDKN2A/B homozygous deletion status of IDH-mutant astrocytoma. STUDY TYPE: Retrospective. POPULATION: Two hundred fifty-one patients: 107 patients with CDKN2A/B homozygous deletion and 144 patients without CDKN2A/B homozygous deletion. FIELD STRENGTH/SEQUENCE: 3.0 T/1.5 T: Contrast-enhanced T1-weighted spin-echo inversion recovery sequence (CE-T1WI) and T2-weighted fluid-attenuation spin-echo inversion recovery sequence (T2FLAIR). ASSESSMENT: A total of 1106 radiomics and 1000 deep learning features extracted from CE-T1WI and T2FLAIR were used to develop models to discriminate the CDKN2A/B homozygous deletion status. Radiomics models, deep learning-based radiomics (DLR) models and the final integrated model combining radiomics features with deep learning features were developed and compared their preoperative discrimination performance. STATISTICAL TESTING: Pearson chi-square test and Mann Whitney U test were used for assessing the statistical differences in patients' clinical characteristics. The Delong test compared the statistical differences of receiver operating characteristic (ROC) curves and area under the curve (AUC) of different models. The significance threshold is P < 0.05. RESULTS: The final combined model (training AUC = 0.966; validation AUC = 0.935; test group: AUC = 0.943) outperformed the optimal models based on only radiomics or DLR features (training: AUC = 0.916 and 0.952; validation: AUC = 0.886 and 0.912; test group: AUC = 0.862 and 0.902). DATA CONCLUSION: Whether based on a single sequence or a combination of two sequences, radiomics and DLR models have achieved promising performance in assessing CDKN2A/B homozygous deletion status. However, the final model combining both deep learning and radiomics features from CE-T1WI and T2FLAIR outperformed the optimal radiomics or DLR model. EVIDENCE LEVEL: 4 TECHNICAL EFFICACY: Stage 2.

Feasibility and effectiveness of automatic deep learning network and radiomics models for differentiating tumor stroma ratio in pancreatic ductal adenocarcinoma.

OBJECTIVE: This study aims to compare the feasibility and effectiveness of automatic deep learning network and radiomics models in differentiating low tumor stroma ratio (TSR) from high TSR in pancreatic ductal adenocarcinoma (PDAC). METHODS: A retrospective analysis was conducted on a total of 207 PDAC patients from three centers (training cohort: n = 160; test cohort: n = 47). TSR was assessed on hematoxylin and eosin-stained specimens by experienced pathologists and divided as low TSR and high TSR. Deep learning and radiomics models were developed including ShuffulNetV2, Xception, MobileNetV3, ResNet18, support vector machine (SVM), k-nearest neighbor (KNN), random forest (RF), and logistic regression (LR). Additionally, the clinical models were constructed through univariate and multivariate logistic regression. Kaplan-Meier survival analysis and log-rank tests were conducted to compare the overall survival time between different TSR groups. RESULTS: To differentiate low TSR from high TSR, the deep learning models based on ShuffulNetV2, Xception, MobileNetV3, and ResNet18 achieved AUCs of 0.846, 0.924, 0.930, and 0.941, respectively, outperforming the radiomics models based on SVM, KNN, RF, and LR with AUCs of 0.739, 0.717, 0.763, and 0.756, respectively. Resnet 18 achieved the best predictive performance. The clinical model based on T stage alone performed worse than deep learning models and radiomics models. The survival analysis based on 142 of the 207 patients demonstrated that patients with low TSR had longer overall survival. CONCLUSIONS: Deep learning models demonstrate feasibility and superiority over radiomics in differentiating TSR in PDAC. The tumor stroma ratio in the PDAC microenvironment plays a significant role in determining prognosis. CRITICAL RELEVANCE STATEMENT: The objective was to compare the feasibility and effectiveness of automatic deep learning networks and radiomics models in identifying the tumor-stroma ratio in pancreatic ductal adenocarcinoma. Our findings demonstrate deep learning models exhibited superior performance compared to traditional radiomics models. KEY POINTS: • Deep learning demonstrates better performance than radiomics in differentiating tumor-stroma ratio in pancreatic ductal adenocarcinoma. • The tumor-stroma ratio in the pancreatic ductal adenocarcinoma microenvironment plays a protective role in prognosis. • Preoperative prediction of tumor-stroma ratio contributes to clinical decision-making and guiding precise medicine.

Automatic Evaluating of Multi-Phase Cranial CTA Collateral Circulation Based on Feature Fusion Attention Network Model.

Stroke is one of the main causes of disability and death, and it can be divided into hemorrhagic stroke and ischemic stroke. Ischemic stroke is more common, and about 8 out of 10 stroke patients suffer from ischemic stroke. In clinical practice, doctors diagnose stroke by using computed tomography angiography (CTA) image to accurately evaluate the collateral circulation in stroke patients. This imaging information is of great significance in assisting doctors to determine the patient's treatment plan and prognosis. Currently, great progress has been made in the field of computer-aided diagnosis technology in medicine by using artificial intelligence. However, in related research based on deep learning algorithms, researchers usually only use single-phase data for training, lacking the temporal dimension information of multi-phase image data. This makes it difficult for the model to learn more comprehensive and effective collateral circulation feature representation, thereby limiting its performance. Therefore, combining data for training is expected to improve the accuracy and reliability of collateral circulation evaluation. In this study, we propose an effective hybrid mechanism to assist the feature encoding network in evaluating the degree of collateral circulation in the brain. By using a hybrid attention mechanism, additional guidance and regularization are provided to enhance the collateral circulation feature representation across multiple stages. Time dimension information is added to the input, and multiple feature-level fusion modules are designed in the multi-branch network. The first fusion module in the single-stage feature extraction network completes the fusion of deep and shallow vessel features in the single-branch network, followed by the multi-stage network feature fusion module, which achieves feature fusion for four stages. Tested on a dataset of multi-phase cranial CTA images, the accuracy rate exceeding 90.43%. The experimental results demonstrate that the addition of these modules can fully explore collateral vessel features, improve feature expression capabilities, and optimize the performance of deep learning network model.

Multicenter clinical radiomics-integrated model based on [18F]FDG PET and multi-modal MRI predict ATRX mutation status in IDH-mutant lower-grade gliomas.

OBJECTIVES: To develop a clinical radiomics-integrated model based on 18 F-fluorodeoxyglucose positron emission tomography ([18F]FDG PET) and multi-modal MRI for predicting alpha thalassemia/mental retardation X-linked (ATRX) mutation status of IDH-mutant lower-grade gliomas (LGGs). METHODS: One hundred and two patients (47 ATRX mutant-type, 55 ATRX wild-type) diagnosed with IDH-mutant LGGs (CNS WHO grades 1 and 2) were retrospectively enrolled. A total of 5540 radiomics features were extracted from structural MR (sMR) images (contrast-enhanced T1-weighted imaging, CE-T1WI; T2-weighted imaging, and T2WI), functional MR (fMR) images (apparent diffusion coefficient, ADC; cerebral blood volume, CBV), and metabolic PET images ([18F]FDG PET). The random forest algorithm was used to establish a clinical radiomics-integrated model, integrating the optimal multi-modal radiomics model with three clinical parameters. The predictive effectiveness of the models was evaluated by receiver operating characteristic (ROC) and decision curve analysis (DCA). RESULTS: The optimal multi-modal model incorporated sMR (CE-T1WI), fMR (ADC), and metabolic ([18F]FDG) images ([18F]FDG PET+ADC+ CE-T1WI) with the area under curves (AUCs) in the training and test groups of 0.971 and 0.962, respectively. The clinical radiomics-integrated model, incorporating [18F]FDG PET+ADC+CE-T1WI, three clinical parameters (KPS, SFSD, and ATGR), showed the best predictive effectiveness in the training and test groups (0.987 and 0.975, respectively). CONCLUSIONS: The clinical radiomics-integrated model with metabolic, structural, and functional information based on [18F]FDG PET and multi-modal MRI achieved promising performance for predicting the ATRX mutation status of IDH-mutant LGGs. KEY POINTS: • The clinical radiomics-integrated model based on [18F]FDG PET and multi-modal MRI achieved promising performance for predicting ATRX mutation status in LGGs. • The study investigated the value of multicenter clinical radiomics-integrated model based on [18F]FDG PET and multi-modal MRI in LGGs regarding ATRX mutation status prediction. • The integrated model provided structural, functional, and metabolic information simultaneously and demonstrated with satisfactory calibration and discrimination in the training and test groups (0.987 and 0.975, respectively).

Optimizing DUS testing for Chimonanthus praecox using feature selection based on a genetic algorithm.

Chimonanthus praecox is a famous traditional flower in China with high ornamental value. It has numerous varieties, yet its classification is highly disorganized. The distinctness, uniformity, and stability (DUS) test enables the classification and nomenclature of various species; thus, it can be used to classify the Chimonanthus varieties. In this study, flower traits were quantified using an automatic system based on pattern recognition instead of traditional manual measurement to improve the efficiency of DUS testing. A total of 42 features were quantified, including 28 features in the DUS guidelines and 14 new features proposed in this study. Eight algorithms were used to classify wintersweet, and the random forest (RF) algorithm performed the best when all features were used. The classification accuracy of the outer perianth was the highest when the features of the different parts were used for classification. A genetic algorithm was used as the feature selection algorithm to select a set of 22 reduced core features and improve the accuracy and efficiency of the classification. Using the core feature set, the classification accuracy of the RF model improved to 99.13%. Finally, K-means was used to construct a pedigree cluster tree of 23 varieties of wintersweet; evidently, wintersweet was clustered into a single class, which can be the basis for further study of genetic relationships among varieties. This study provides a novel method for DUS detection, variety identification, and pedigree analysis.

One shot ancient character recognition with siamese similarity network.

Ancient character recognition is not only important for the study and understanding of ancient history but also has a profound impact on the inheritance and development of national culture. In order to reduce the study of difficult professional knowledge of ancient characters, and meanwhile overcome the lack of data, class imbalance, diversification of glyphs, and open set recognition problems in ancient characters, we propose a Siamese similarity network based on a similarity learning method to directly learn input similarity and then apply the trained model to establish one shot classification task for recognition. Multi-scale fusion backbone structure and embedded structure are proposed in the network to improve the model's ability to extract features. We also propose the soft similarity contrast loss function for the first time, which ensures the optimization of similar images with higher similarity and different classes of images with greater differences while reducing the over-optimization of back-propagation leading to model overfitting. Specially, we propose a cumulative class prototype based on our network to solve the deviation problem of the mean class prototype and obtain a good class representation. Since new ancient characters can still be found in reality, our model has the ability to reject unknown categories while identifying new ones. A large number of experiments show that our proposed method has achieved high-efficiency discriminative performance and obtained the best performance over the methods of traditional deep learning and other classic one-shot learning.

Merging Multiphase CTA Images and Training Them Simultaneously with a Deep Learning Algorithm Could Improve the Efficacy of AI Models for Lateral Circulation Assessment in Ischemic Stroke.

We aimed to build a deep learning-based, objective, fast, and accurate collateral circulation assessment model. We included 92 patients who had suffered acute ischemic stroke (AIS) with large vessel occlusion in the anterior circulation in this study, following their admission to our hospital from June 2020 to August 2021. We analyzed their baseline whole-brain four-dimensional computed tomography angiography (4D-CTA)/CT perfusion. The images of the arterial, arteriovenous, venous, and late venous phases were extracted from 4D-CTA according to the perfusion time-density curve. The subtraction images of each phase were created by subtracting the non-contrast CT. Each patient was marked as having good or poor collateral circulation. Based on the ResNet34 classification network, we developed a single-image input and a multi-image input network for binary classification of collateral circulation. The training and test sets included 65 and 27 patients, respectively, and Monte Carlo cross-validation was employed for five iterations. The network performance was evaluated based on its precision, accuracy, recall, F1-score, and AUC. All the five performance indicators of the single-image input model were higher than those of the other model. The single-image input processing network, combining multiphase CTA images, can better classify AIS collateral circulation. This automated collateral assessment tool could help to streamline clinical workflows, and screen patients for reperfusion therapy.

Clinical Variables, Deep Learning and Radiomics Features Help Predict the Prognosis of Adult Anti-N-methyl-D-aspartate Receptor Encephalitis Early: A Two-Center Study in Southwest China.

Objective: To develop a fusion model combining clinical variables, deep learning (DL), and radiomics features to predict the functional outcomes early in patients with adult anti-N-methyl-D-aspartate receptor (NMDAR) encephalitis in Southwest China. Methods: From January 2012, a two-center study of anti-NMDAR encephalitis was initiated to collect clinical and MRI data from acute patients in Southwest China. Two experienced neurologists independently assessed the patients' prognosis at 24 moths based on the modified Rankin Scale (mRS) (good outcome defined as mRS 0-2; bad outcome defined as mRS 3-6). Risk factors influencing the prognosis of patients with acute anti-NMDAR encephalitis were investigated using clinical data. Five DL and radiomics models trained with four single or combined four MRI sequences (T1-weighted imaging, T2-weighted imaging, fluid-attenuated inversion recovery imaging and diffusion weighted imaging) and a clinical model were developed to predict the prognosis of anti-NMDAR encephalitis. A fusion model combing a clinical model and two machine learning-based models was built. The performances of the fusion model, clinical model, DL-based models and radiomics-based models were compared using the area under the receiver operating characteristic curve (AUC) and accuracy and then assessed by paired t-tests (P < 0.05 was considered significant). Results: The fusion model achieved the significantly greatest predictive performance in the internal test dataset with an AUC of 0.963 [95% CI: (0.874-0.999)], and also significantly exhibited an equally good performance in the external validation dataset, with an AUC of 0.927 [95% CI: (0.688-0.975)]. The radiomics_combined model (AUC: 0.889; accuracy: 0.857) provided significantly superior predictive performance than the DL_combined (AUC: 0.845; accuracy: 0.857) and clinical models (AUC: 0.840; accuracy: 0.905), whereas the clinical model showed significantly higher accuracy. Compared with all single-sequence models, the DL_combined model and the radiomics_combined model had significantly greater AUCs and accuracies. Conclusions: The fusion model combining clinical variables and machine learning-based models may have early predictive value for poor outcomes associated with anti-NMDAR encephalitis.

Differentiating solitary brain metastases from glioblastoma by radiomics features derived from MRI and 18F-FDG-PET and the combined application of multiple models.

This study aimed to explore the ability of radiomics derived from both MRI and 18F-fluorodeoxyglucose positron emission tomography (18F-FDG-PET) images to differentiate glioblastoma (GBM) from solitary brain metastases (SBM) and to investigate the combined application of multiple models. The imaging data of 100 patients with brain tumours (50 GBMs and 50 SBMs) were retrospectively analysed. Three model sets were built on MRI, 18F-FDG-PET, and MRI combined with 18F-FDG-PET using five feature selection methods and five classification algorithms. The model set with the highest average AUC value was selected, in which some models were selected and divided into Groups A, B, and C. Individual and joint voting predictions were performed in each group for the entire data. The model set based on MRI combined with 18F-FDG-PET had the highest average AUC compared with isolated MRI or 18F-FDG-PET. Joint voting prediction showed better performance than the individual prediction when all models reached an agreement. In conclusion, radiomics derived from MRI and 18F-FDG-PET could help differentiate GBM from SBM preoperatively. The combined application of multiple models can provide greater benefits.

An Integrated Radiomics Model Incorporating Diffusion-Weighted Imaging and 18F-FDG PET Imaging Improves the Performance of Differentiating Glioblastoma From Solitary Brain Metastases.

BACKGROUND: The effectiveness of conventional MRI (cMRI)-based radiomics in differentiating glioblastoma (GBM) from solitary brain metastases (SBM) is not satisfactory enough. Therefore, we aimed to develop an integrated radiomics model to improve the performance of differentiating GBM from SBM. METHODS: One hundred patients with solitary brain tumors (50 with GBM, 50 with SBM) were retrospectively enrolled and randomly assigned to the training set (n = 80) or validation set (n = 20). A total of 4,424 radiomic features were obtained from contrast-enhanced T1-weighted imaging (CE-T1WI) with the contrast-enhancing and peri-enhancing edema region, T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI)-derived apparent diffusion coefficient (ADC), and 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET) images. The partial least squares (PLS) regression with fivefold cross-validation is used to analyze the correlation between different radiomic features and different modalities. The cross-validity analysis was performed to judge whether a new principal component or a new feature dimension can significantly improve the final prediction effect. The principal components with effective interpretation in all radiomic features were projected to a low-dimensional space (2D in this study). The effective features of the new projection mapping were then sent to the random forest classifier to predict the results. The performance of differentiating GBM from SBM was compared between the integrated radiomics model and other radiomics models or nonradiomics methods using the area under the receiver operating characteristics curve (AUC). RESULTS: Through the cross-validity analysis of partial least squares, hundreds of radiomic features were projected into a new two-dimensional space to complete the construction of radiomics model. Compared with the combined radiomics model using DWI + 18F-FDG PET (AUC = 0.93, p = 0.014), cMRI + DWI (AUC = 0.89, p = 0.011), cMRI + 8F-FDG PET (AUC = 0.91, p = 0.015), and single radiomics model using cMRI (AUC = 0.85, p = 0.018), DWI (AUC = 0.84, p = 0.017), and 18F-FDG PET (AUC = 0.85, p = 0.421), the integrated radiomics model (AUC = 0.98) showed more efficient diagnostic performance. The integrated radiomics model (AUC = 0.98) also showed significantly better performance than any single ADC, SUV, or TBR parameter (AUC = 0.57-0.71, p < 0.05). The integrated radiomics model showed better performance in the training (AUC = 0.98) and validation (AUC = 0.93) sets than any other models and methods, demonstrating robustness. CONCLUSIONS: We developed an integrated radiomics model incorporating DWI and 18F-FDG PET, which improved the performance of differentiating GBM from SBM greatly.

A Classification Method of Oracle Materials Based on Local Convolutional Neural Network Framework.

The classification of materials of oracle bone is one of the most basic aspects for oracle bone morphology. However, the classification method depending on experts' experience requires long-term learning and accumulation for professional knowledge. This article presents a multiregional convolutional neural network to classify the rubbings of oracle bones. First, we detected the "shield grain" and "tooth grain" on the oracle bone rubbings, then complete the division of multiple areas on an image of oracle bone. Second, the convolutional neural network is used to extract the features of each region and we complete the fusion of multiple local features. Finally, the classification of tortoise shell and animal bone was realized. Utilizing the image of oracle bone provided by experts, we conducted an experiment; the result show our method has better classification accuracy. It has made contributions to the progress of the study of oracle bone morphology.