Classification of piano performers with deep learning models

PurposeTo conduct a head-to-head comparison between deep learning (DL) and radiomics models across institutions for predicting microvascular invasion (MVI) in hepatocellular carcinoma (HCC) and to investigate the model robustness and generalizability through rigorous internal and external validation.MethodsThis retrospective study included 2304 preoperative images of 576 HCC lesions from two centers, with MVI status determined by postoperative histopathology. We developed DL and radiomics models for predicting the presence of MVI using B-mode ultrasound, contrast-enhanced ultrasound (CEUS) at the arterial, portal, and delayed phases, and a combined modality (B + CEUS). For radiomics, we constructed models with enlarged vs. original regions of interest (ROIs). A cross-validation approach was performed by training models on one center’s dataset and validating the other, and vice versa. This allowed assessment of the validity of different ultrasound modalities and the cross-center robustness of the models. The optimal model combined with alpha-fetoprotein (AFP) was also validated. The head-to-head comparison was based on the area under the receiver operating characteristic curve (AUC).ResultsThirteen DL models and 25 radiomics models using different ultrasound modalities were constructed and compared. B + CEUS was the optimal modality for both DL and radiomics models. The DL model achieved AUCs of 0.802–0.818 internally and 0.667–0.688 externally across the two centers, whereas radiomics achieved AUCs of 0.749–0.869 internally and 0.646–0.697 externally. The radiomics models showed overall improvement with enlarged ROIs (P < 0.05 for both CEUS and B + CEUS modalities). The DL models showed good cross-institutional robustness (P > 0.05 for all modalities, 1.6–2.1% differences in AUC for the optimal modality), whereas the radiomics models had relatively limited robustness across the two centers (12% drop-off in AUC for the optimal modality). Adding AFP improved the DL models (P < 0.05 externally) and well maintained the robustness, but did not benefit the radiomics model (P > 0.05).ConclusionCross-institutional validation indicated that DL demonstrated better robustness than radiomics for preoperative MVI prediction in patients with HCC, representing a promising solution to non-standardized ultrasound examination procedures.

We aimed at comparing intratumoral and peritumoral deep learning, radiomics, and fusion models in predicting KRAS mutations in rectal cancer using endorectal ultrasound imaging. This study included 304 patients with rectal cancer from Fujian Medical University Union Hospital. The patients were randomly divided into a training group (213 patients) and a test group (91 patients) at a 7:3 ratio. Radiomics and deep learning models were established using primary tumor and peritumoral images. In the optimally performing regions-of-interest, two fusion strategies, a feature-based and a decision-based model, were employed to build the fusion models. The Shapley additive explanation (SHAP) method was used to evaluate the significance of features in the optimal radiomics, deep learning, and fusion models. The performance of each model was assessed using the area under the receiver operating characteristic curve (AUC) and decision curve analysis (DCA). In the test cohort, both the radiomics and deep learning models exhibited optimal performance with a 10-pixel patch extension, yielding AUC values of 0.824 and 0.856, respectively. The feature-based DLRexpand10_FB model attained the highest AUC (0.896) across all study sets. In addition, the DLRexpand10_FB model demonstrated excellent sensitivity, specificity, and DCA. SHAP analysis underscored the deep learning feature (DL_1) as the most significant factor in the hybrid model. The feature-based fusion model DLRexpand10_FB can be employed to predict KRAS gene mutations based on pretreatment endorectal ultrasound images of rectal cancer. The integration of peritumoral regions enhanced the predictive performance of both the radiomics and deep learning models.

Deep learning breast cancer risk models demonstrate improved accuracy compared with traditional risk models but have not been prospectively tested. We compared the accuracy of a deep learning risk score derived from the patient's prior mammogram to traditional risk scores to prospectively identify patients with cancer in a cohort due for screening. We collected data on 119 139 bilateral screening mammograms in 57 617 consecutive patients screened at 5 facilities between September 18, 2017, and February 1, 2021. Patient demographics were retrieved from electronic medical records, cancer outcomes determined through regional tumor registry linkage, and comparisons made across risk models using Wilcoxon and Pearson χ2 2-sided tests. Deep learning, Tyrer-Cuzick, and National Cancer Institute Breast Cancer Risk Assessment Tool (NCI BCRAT) risk models were compared with respect to performance metrics and area under the receiver operating characteristic curves. Cancers detected per thousand patients screened were higher in patients at increased risk by the deep learning model (8.6, 95% confidence interval [CI] = 7.9 to 9.4) compared with Tyrer-Cuzick (4.4, 95% CI = 3.9 to 4.9) and NCI BCRAT (3.8, 95% CI = 3.3 to 4.3) models (P &lt; .001). Area under the receiver operating characteristic curves of the deep learning model (0.68, 95% CI = 0.66 to 0.70) was higher compared with Tyrer-Cuzick (0.57, 95% CI = 0.54 to 0.60) and NCI BCRAT (0.57, 95% CI = 0.54 to 0.60) models. Simulated screening of the top 50th percentile risk by the deep learning model captured statistically significantly more patients with cancer compared with Tyrer-Cuzick and NCI BCRAT models (P &lt; .001). A deep learning model to assess breast cancer risk can support feasible and effective risk-based screening and is superior to traditional models to identify patients destined to develop cancer in large screening cohorts.

Electricity load forecasting is a crucial undertaking within all the deregulated markets globally. Among the research challenges on a global scale, the investigation of deep transfer learning (DTL) in the field of electricity load forecasting represents a fundamental effort that can inform artificial intelligence applications in general. In this paper, a comprehensive study is reported regarding day-ahead electricity load forecasting. For this purpose, three sequence-to-sequence (Seq2seq) deep learning (DL) models are used, namely the multilayer perceptron (MLP), the convolutional neural network (CNN) and the ensemble learning model (ELM), which consists of the weighted combination of the outputs of MLP and CNN models. Also, the study focuses on the development of different forecasting strategies based on DTL, emphasizing the way the datasets are trained and fine-tuned for higher forecasting accuracy. In order to implement the forecasting strategies using deep learning models, load datasets from three Greek islands, Rhodes, Lesvos, and Chios, are used. The main purpose is to apply DTL for day-ahead predictions (1–24 h) for each month of the year for the Chios dataset after training and fine-tuning the models using the datasets of the three islands in various combinations. Four DTL strategies are illustrated. In the first strategy (DTL Case 1), each of the three DL models is trained using only the Lesvos dataset, while fine-tuning is performed on the dataset of Chios island, in order to create day-ahead predictions for the Chios load. In the second strategy (DTL Case 2), data from both Lesvos and Rhodes concurrently are used for the DL model training period, and fine-tuning is performed on the data from Chios. The third DTL strategy (DTL Case 3) involves the training of the DL models using the Lesvos dataset, and the testing period is performed directly on the Chios dataset without fine-tuning. The fourth strategy is a multi-task deep learning (MTDL) approach, which has been extensively studied in recent years. In MTDL, the three DL models are trained simultaneously on all three datasets and the final predictions are made on the unknown part of the dataset of Chios. The results obtained demonstrate that DTL can be applied with high efficiency for day-ahead load forecasting. Specifically, DTL Case 1 and 2 outperformed MTDL in terms of load prediction accuracy. Regarding the DL models, all three exhibit very high prediction accuracy, especially in the two cases with fine-tuning. The ELM excels compared to the single models. More specifically, for conducting day-ahead predictions, it is concluded that the MLP model presents the best monthly forecasts with MAPE values of 6.24% and 6.01% for the first two cases, the CNN model presents the best monthly forecasts with MAPE values of 5.57% and 5.60%, respectively, and the ELM model achieves the best monthly forecasts with MAPE values of 5.29% and 5.31%, respectively, indicating the very high accuracy it can achieve.

This paper takes the problems and ways of diversified curriculum construction of vocal music education in colleges and universities as the entry point and constructs a deep blended learning model of vocal music diversified curriculum based on the deep learning model. The deep evolutionary knowledge tracking model is constructed by combining the self-attention mechanism embedding model with vocal music knowledge evolution based on Transformer. To verify the effectiveness of the deep blended learning model constructed in this paper, model comparison experiments and empirical analysis were conducted. The results show that compared with the SAKT model, the DBL model in this paper has a 5.19% improvement in the average ROC-AUC value. The vocal posttest scores of the students in the experimental class were 5.09 points higher than those of the control class, with a two-tailed significance value of 0.027, which is less than 0.05 and a significant difference. This indicates that the deep learning model can effectively promote the innovative design of diversified curricula for vocal music education in colleges and universities and enhance students’ vocal music learning ability.

Objective.Functional brain network (FBN) methods are commonly integrated with deep learning (DL) models for EEG analysis. Typically, an FBN is constructed to extract features from EEG data, which are then fed into a DL model for further analysis. Beyond this two-step approach, there is potential to embed FBN construction directly within DL models as a feature extraction module, enabling the models to learn EEG representations end-to-end while incorporating insights from FBNs. However, a critical prerequisite is whether DL models can effectively learn the FBN construction process.Approach.To address this, we propose using DL models to learn FBN matrices derived from EEG data. The ability of DL models to accurately reproduce these matrices would validate their capacity to learn the FBN construction process. This approach is tested on two publicly available EEG datasets, utilizing seven DL models to learn four representative FBN matrices. Model performance is assessed through mean squared error (MSE), Pearson correlation coefficient (Corr), and concordance correlation coefficient (CCC) between predicted and actual matrices.Main results.The results show that DL models achieve low MSE and relatively high Corr and CCC values when learning the Coherence network. Visualizations of predicted and error matrices reveal that while DL models capture the general structure of all four FBNs, certain regions remain difficult to model accurately. Additionally, a pairedt-test comparing global efficiency and nodal degree between predicted and actual networks indicates that most predicted networks significantly differ from the actual networks (p<0.05).Significance.These findings suggest that while DL models can learn the connectivity relationships of certain FBNs, they struggle to capture the intrinsic topological structures. This highlights the irreplaceability of traditional FBN methods in EEG analysis and underscores the need for hybrid strategies that combine FBN methods with DL models for a more comprehensive analysis.

To develop and evaluate static ultrasound images of ovaries-based radiomics and deep learning models for predicting antral follicle count (AFC) and anti-Müllerian hormone (AMH). A retrospective analysis was performed on 395 infertility patients who completed AFC through transvaginal ultrasonography and AMH test in two branches of the hospital. The data set was divided into training cohort (n = 210, Branch A), internal test cohort (n = 91, Branch A), and external test cohort (n = 94, Branch B). The AFC results of static ultrasound images (AFCstatic) and dynamic scan (AFCdynamic) were analysed for correlation and consistency. The radiomics, deep learning and combined models (Rad-DL model) were developed to quantitatively predict AFC and AMH respectively, and the ordered 3-class classification (AFCdynamic≤5,6≤AFCdynamic≤19 and 20≤AFCdynamic) prediction was further carried out. There was a high degree of correlation and consistency between AFCstatic and AFCdynamic in both branches. In the quantitative prediction of AFCdynamic, the R2 of the radiomics, deep learning and Rad-DL models in the external test cohort were 0.572, 0.541, 0.583, respectively. In the quantitative prediction of AMH, the R2 of the radiomics, deep learning and Rad-DL models in the external test cohort were 0.420, 0.352, 0.509, respectively. For the ordered 3-class classification prediction, the radiomics, deep learning and Rad-DL models demonstrated a good discrimination performance with overall AUCs of 0.816 (95%CI, 0.747–0.887), 0.843 (95%CI, 0.772–0.901), and 0.881 (95%CI, 0.816–0.931) in the external test cohort, respectively. Radiomics and deep learning models developed by using static ultrasound images of ovaries may quantitatively predict AFCdynamic to a certain extent, and accurately screen infertility patients with different ovarian reserve levels. This study lays the foundation for further break through the rigid antral follicle count and explore an intelligent non-invasive ovarian reserve evaluation method.

Introduction: Rhegmatogenous retinal detachment (RRD) is one of the most common fundus diseases. Many rural areas of China have few ophthalmologists, and ophthalmologic ultrasound examination is of great significance for remote diagnosis of RRD. Therefore, this study aimed to develop and evaluate a deep learning (DL) model, to be used for automated RRD diagnosis based on ophthalmologic ultrasound images, in order to support timely diagnosis of RRD in rural and remote areas. Methods: A total of 6,000 ophthalmologic ultrasound images from 1,645 participants were used to train and verify the DL model. A total of 5,000 images were used for training and validating DL models, and an independent testing set of 1,000 images was used to test the performance of eight DL models trained using four different DL model architectures (fully connected neural network, LeNet5, AlexNet, and VGG16) and two preprocessing techniques (original, original image augmented). Receiver operating characteristic (ROC) curves were used to analyze their performance. Heatmaps were generated to visualize the process of the best DL model in the identification of RRD. Finally, five ophthalmologists were invited to diagnose RRD independently on the same test set of 1,000 images for performance comparison with the best DL model. Results: The best DL model for identifying RRD achieved an area under the ROC curve (AUC) of 0.998 with a sensitivity and specificity of 99.2% and 99.8%, respectively. The best preprocessing method in each model architecture was the application of original image augmentation (average AUC = 0.982). The best model architecture in each preprocessing method was VGG16 (average AUC = 0.998). Conclusion: The best DL model determined in this study has higher accuracy, sensitivity, and specificity than the ophthalmologists’ diagnosis in identifying RRD based on ophthalmologic ultrasound images. This model may provide support for timely diagnosis in locations without access to ophthalmologic care.

COVID-19 is highly infectious and causes acute respiratory disease. Machine learning (ML) and deep learning (DL) models are vital in detecting disease from computerized chest tomography (CT) scans. The DL models outperformed the ML models. For COVID-19 detection from CT scan images, DL models are used as end-to-end models. Thus, the performance of the model is evaluated for the quality of the extracted feature and classification accuracy. There are four contributions included in this work. First, this research is motivated by studying the quality of the extracted feature from the DL by feeding these extracted to an ML model. In other words, we proposed comparing the end-to-end DL model performance against the approach of using DL for feature extraction and ML for the classification of COVID-19 CT scan images. Second, we proposed studying the effect of fusing extracted features from image descriptors, e.g., Scale-Invariant Feature Transform (SIFT), with extracted features from DL models. Third, we proposed a new Convolutional Neural Network (CNN) to be trained from scratch and then compared to the deep transfer learning on the same classification problem. Finally, we studied the performance gap between classic ML models against ensemble learning models. The proposed framework is evaluated using a CT dataset, where the obtained results are evaluated using five different metrics The obtained results revealed that using the proposed CNN model is better than using the well-known DL model for the purpose of feature extraction. Moreover, using a DL model for feature extraction and an ML model for the classification task achieved better results in comparison to using an end-to-end DL model for detecting COVID-19 CT scan images. Of note, the accuracy rate of the former method improved by using ensemble learning models instead of the classic ML models. The proposed method achieved the best accuracy rate of 99.39%.

The past few years have seen a surge of applying Deep Learning (DL) models for a wide array of tasks such as image classification, object detection, machine translation, etc. While DL models provide an opportunity to solve otherwise intractable tasks, their adoption relies on them being optimized to meet latency and resource requirements. Benchmarking is a key step in this process but has been hampered in part due to the lack of representative and up-to-date benchmarking suites. This is exacerbated by the fast-evolving pace of DL models. This paper proposes DLBricks, a composable benchmark generation design that reduces the effort of developing, maintaining, and running DL benchmarks on CPUs. DLBricks decomposes DL models into a set of unique runnable networks and constructs the original model's performance using the performance of the generated benchmarks. DLBricks leverages two key observations: DL layers are the performance building blocks of DL models and layers are extensively repeated within and across DL models. Since benchmarks are generated automatically and the benchmarking time is minimized, DLBricks can keep up-to-date with the latest proposed models, relieving the pressure of selecting representative DL models. Moreover, DLBricks allows users to represent proprietary models within benchmark suites. We evaluate DLBricks using $50$ MXNet models spanning $5$ DL tasks on $4$ representative CPU systems. We show that DLBricks provides an accurate performance estimate for the DL models and reduces the benchmarking time across systems (e.g. within $95\%$ accuracy and up to $4.4\times$ benchmarking time speedup on Amazon EC2 c5.xlarge).

Manufacturing process of Lithium-ion battery electrode has a direct impact on the resulting practical properties of the cell such as durability, safety and overall performance. In this scenario, together with experimental efforts to understand the correlation between manufacturing process parameters and final overall electrode performance, computational tools have shown the potential to produce insights on the manufacturing properties interdependencies. The ARTISTIC initiative [1] pioneered the development of a series of computational 3D-resolved physics-based models describing each step of the manufacturing process following a sequential pattern, going from the slurry phase, through the drying phase, to produce a calendered electrode. Such physics-based models have the capability to predict how manufacturing parameters impact the electrode microstructure. At the end of each phase the output microstructure is carefully validated with experimental data, assuring that the model is in good agreement with real properties. For the calendering phase, the Discrete Element Method (DEM) is used for the model simulation.One of the main limitations of the DEM simulations is their high computational cost, preventing their direct application in electrode optimization loops. In this matter, and given the great amount of already validated data that was produced in our previous works by using the ARTISTIC model, we take one step forward in the development of a new generation model employing machine learning (ML) techniques to accelerate the physics-based simulations but keeping their accuracy. In this presentation we report a novel time-dependent deep learning (DL) model of the battery electrodes manufacturing process. The DL model is demonstrated for calendering of nickel manganese cobalt (NMC111) electrodes, and trained with time-series data arising from physics-based DEM simulations coming from the ARTISTIC physics-based modeling pipeline [2]. The innovative DL model presented in this work can predict an electrode microstructure evolution over time at a given compression degree during calendering and provide the associated final relaxed electrode microstructure. Here we used the formulation of 96% AM and 4% CBD as a proof of concept.The DL model predictions are validated by comparing data evaluation metrics (e.g., mean square error (MSE) and R2 score) and electrode functional metrics (contact surface area, porosity, diffusivity, and tortuosity factor), showing very good accuracy with respect to the DEM simulations (all the properties are above 90% accurate when compared to target data). Regarding transport properties and given the high accuracy of the DL model to predict tortuosity factor and porosity values we can conclude that the model correctly predicts the effect of the calendering on the transport properties since tortuosity factor and porosity has a high impact on the effective transport properties of Li-ions in the electrolyte. The DL model can remarkably capture the elastic recovery of the electrode upon compression (spring-back phenomenon) and the main 3D electrode microstructure features without using the functional descriptors during its training. In terms of computational cost, the DL model performs a timestep in 15 seconds (wall time) while the DEM model performs an equivalent time step in ∼47 minutes (wall time). Given the DL model significant lower computational cost than the DEM simulations, it paves the way toward quasi-real-time optimization loops of the 3D electrode architecture predicting the calendering conditions to adopt in order to obtain the desired electrode performance. Moreover, we consider that our model has tremendous potential to grow in several aspects such as accuracy and predictability. Therefore, our incoming efforts will be focused on continuing the development of the model and testing it for other formulations and in other manufacturing stages.[1] ARTISTIC Project website. https://www.erc-artistic.eu.[2] D. E. Galvez-Aranda, T. L. Dinh, U. Vijay, F. M. Zanotto, A. A. Franco, Time-Dependent Deep Learning Manufacturing Process Model for Battery Electrode Microstructure Prediction. Adv. Energy Mater. 2024, 2400376. https://doi.org/10.1002/aenm.202400376 Figure 1

PurposeTo develop generative adversarial networks (GANs) that synthesize realistic anterior segment optical coherence tomography (AS-OCT) images and evaluate deep learning (DL) models that are trained on real and synthetic datasets for detecting angle closure.MethodsThe GAN architecture was adopted and trained on the dataset with AS-OCT images collected from the Joint Shantou International Eye Center of Shantou University and the Chinese University of Hong Kong, synthesizing open- and closed-angle AS-OCT images. A visual Turing test with two glaucoma specialists was performed to assess the image quality of real and synthetic images. DL models, trained on either real or synthetic datasets, were developed. Using the clinicians’ grading of the AS-OCT images as the reference standard, we compared the diagnostic performance of open-angle vs. closed-angle detection of DL models and the AS-OCT parameter, defined as a trabecular-iris space area 750 µm anterior to the scleral spur (TISA750), in a small independent validation dataset.ResultsThe GAN training included 28,643 AS-OCT anterior chamber angle (ACA) images. The real and synthetic datasets for DL model training have an equal distribution of open- and closed-angle images (all with 10,000 images each). The independent validation dataset included 238 open-angle and 243 closed-angle AS-OCT ACA images. The image quality of real versus synthetic AS-OCT images was similar, as assessed by the two glaucoma specialists, except for the scleral spur visibility. For the independent validation dataset, both DL models achieved higher areas under the curve compared with TISA750. Two DL models had areas under the curve of 0.97 (95% confidence interval, 0.96–0.99) and 0.94 (95% confidence interval, 0.92–0.96).ConclusionsThe GAN synthetic AS-OCT images appeared to be of good quality, according to the glaucoma specialists. The DL models, trained on all-synthetic AS-OCT images, can achieve high diagnostic performance.Translational RelevanceThe GANs can generate realistic AS-OCT images, which can also be used to train DL models.

This study sought to develop an innovative body composition (BC)-based deep learning (DL) model to precisely evaluate survival in gastric cancer (GC) patients undergoing neoadjuvant treatment (NT). This retrospective study included GC patients undergoing NT from two centers. CT images both pre-NT and post-NT were preprocessed, focusing on the automatic segmentation of subcutaneous fat, visceral fat, and skeletal muscle regions using TotalSegmentator. Delta Radiomics features were extracted using Pyradiomics. After feature fusion and selection, the optimal model is Naive Bayes (Rad model). A hybrid DL model was developed by combining ResNet18 and Transformer networks for feature extraction. The Clinic_Rad_DL model was constructed by combining clinical features, radiomic signatures, and DL signatures. The ExtraTree classifier was used for the Clinic_Rad_DL model, while a separate Cox regression model was developed for survival analysis using the same features. A total of 356 patients (mean age, 59 ± 10 years; 264 males [74.2%]) were enrolled and divided into training, validation, and test sets in a 7:2:1 ratio. The DL model outperformed the Rad model. The Clinic_Rad_DL model outperformed both Rad model and DL model, with AUC of 0.915, 0.890, and 0.890 in training, validation, and test sets, respectively. The Cox proportional hazards model showed C-index of 0.806, 0.803, and 0.819, effectively stratifying patients into high- and low-risk groups with significant survival differences. The study developed and validated a BC-based DL model to predict survival in GC patients undergoing NT, offering potential for personalized treatment strategies in clinical practice.

We developed a multimodal ultrasound (US) deep learning (DL) fusion model to automatically classify early fibrosis in patients with chronic kidney disease (CKD). This prospective study included patients with CKD who underwent continuous gray-scale US, superb microvascular imaging, and strain elastography from May to November 2022. According to the pathological tubular atrophy and interstitial fibrosis score, patients were divided into minimal and mild groups (affected area ≤10% and 11 − 25% of the total cortical volume, respectively). The dataset was divided into training (70%) and test (30%) sets. A DL model combining the features of the three US modes was developed to predict early fibrosis in patients with CKD. We compared these findings with the area under the receiver operating characteristic curve (AUC) of the clinical model by analyzing the receiver operating characteristic curve in the test set. The AUC of single-mode DL based on gray-scale US, superb microvascular imaging, and strain elastography was 0.682, 0.745, and 0.648, respectively, while that of the multimodal US DL model was 0.86. The accuracy, specificity, and sensitivity of the multimodal US DL model were 0.779, 0.767, and 0.796, respectively, and the negative and positive predictive values were 0.842 and 0.706, respectively. The AUC of the multimodal US DL model was significantly better than that of the single-mode DL and clinical models. The DL algorithm developed using multimodal US images can effectively predict early fibrosis in patients with CKD with significantly greater accuracy than single-mode DL or clinical models.

BACKGROUND Intracranial germ cell tumors (iGCTs) are divided into germinomas (GEs) and nongerminomatous germ cell tumors (NGGCTs), each having distinct treatments and prognoses. However, predictive biomarkers for differentiation are limited. Here, we aim to develop a deep learning (DL) model that can automatically differentiate GEs from NGGCTs using preoperative magnetic resonance (MR) images and analyzing how our model improved clinical decision making among five physicians. METHODS: This multicenter study included patients with pathology-confirmed iGCTs for the developed set (total n = 280), which was split into training (n = 200) and testing datasets (n= 80), and three independent validation datasets (n = 64), respectively. We developed a DL model based on 3D nnU-Net and preoperative MR images to differentiate GEs from NGGCTs. To evaluate the clinical value of the DL model, five physicians' accuracy and level of consistency in identifying tumor types were compared in the testing and three validation datasets, with and without referring to the DL model’s output. RESULTS The DL model achieved AUCs of 0.950, 0.921, 0.869, and 0.905 in the testing and three validation datasets, respectively, which is superior to using conventional information including MR features and clinical information (AUC=0.768, P=0.009). The physicians’ diagnostic performance improved from an average AUC of 0.695 without reference to the DL output to 0.842 using the DL output in the testing and independent validation datasets (21.2% improvement, P=0.014). CONCLUSIONS We developed and validated a DL model using a large dataset to accurately differentiate GEs from NGGCTs using preoperative MR images.