Neural Network Features Research Articles

A Study on the Differences in Optimized Inputs of Various Data-Driven Methods for Battery Capacity Prediction

As lithium-ion batteries become increasingly popular worldwide, accurately determining their capacity is crucial for various devices that rely on them. Numerous data-driven methods have been applied to evaluate battery-related parameters. In the application of these methods, input features play a critical role. Most researchers often use the same input features to compare the performance of various neural network models. However, because most models are regarded as black-box models, different methods may show different dependencies on specific features given the inherent differences in their internal structures. And the corresponding optimal inputs of different neural network models should be different. Therefore, comparing the differences in optimized input features for different neural networks is essential. This paper extracts 11 types of lithium battery-related health features, and experiments are conducted on two traditional machine learning networks and three advanced deep learning networks in three aspects of input differences. The experiment aims to systematically evaluate how changes in health feature types, dimensions, and data volume affect the performance of different methods and find the optimal input for each method. The results demonstrate that each network has its own optimal input in the aspects of health feature types, dimensions, and data volume. Moreover, under the premise of obtaining more accurate prediction accuracy, different networks have different requirements for input data. Therefore, in the process of using different types of neural networks for battery capacity prediction, it is very important to determine the type, dimension, and number of input health features according to the structure, category, and actual application requirements of the network. Different inputs will lead to larger differences in results. The optimization degree of mean absolute error (MAE) can be improved by 10–50%, and other indicators can also be optimized to varying degrees. Therefore, it is very important to optimize the network in a targeted manner.

Arrears behavior prediction of power users based on BP neural network and multi-scale feature learning: a refined risk assessment framework

This study aims to develop an efficient model to predict the arrears behavior of electricity users by integrating multi-scale feature learning with a backpropagation (BP) neural network. The goal is to provide accurate early warning systems and enhanced risk management tools for power companies. The BP neural network algorithm adjusts weights to minimize prediction errors, while multi-scale feature learning captures the diversity and regularity of user behavior by extracting data from various time dimensions, such as daily, weekly, and monthly intervals. First, electricity usage and weather data from the UMass Smart Dataset are preprocessed, including steps such as data cleaning, standardization, and normalization. Next, features are extracted across three time scales—daily, weekly, and monthly. These features are then input into the BP neural network model using the multi-scale feature learning method. A hierarchical neural network structure is designed to address the characteristics of different scales in distinct layers. Key model parameters are optimized, and a sensitivity analysis is conducted. The experimental results demonstrate that the BP neural network model incorporating multi-scale features outperforms traditional BP neural network models and other control models in several evaluation metrics. Specifically, the Gini coefficient is 0.55, the Kolmogorov-Smirnov statistic is 0.60, the Matthews correlation coefficient is 0.45, and specificity is 0.82. These results indicate that the proposed method offers significant improvements in capturing user behavior patterns and enhancing prediction accuracy. The study concludes that the effective fusion of multi-scale features not only enhances the model’s prediction performance but also strengthens its generalization ability. This method provides an advanced risk management tool for power companies, helping to increase the operational efficiency of smart grids and encouraging further research toward greater intelligence in the field.

A quantitative benchmark of neural network feature selection methods for detecting nonlinear signals

Classification and regression problems can be challenging when the relevant input features are diluted in noisy datasets, in particular when the sample size is limited. Traditional Feature Selection (FS) methods address this issue by relying on some assumptions such as the linear or additive relationship between features. Recently, a proliferation of Deep Learning (DL) models has emerged to tackle both FS and prediction at the same time, allowing non-linear modeling of the selected features. In this study, we systematically assess the performance of DL-based feature selection methods on synthetic datasets of varying complexity, and benchmark their efficacy in uncovering non-linear relationships between features. We also use the same settings to benchmark the reliability of gradient-based feature attribution techniques for Neural Networks (NNs), such as Saliency Maps (SM). A quantitative evaluation of the reliability of these approaches is currently missing. Our analysis indicates that even simple synthetic datasets can significantly challenge most of the DL-based FS and SM methods, while Random Forests, TreeShap, mRMR and LassoNet are the best performing FS methods. Our conclusion is that when quantifying the relevance of a few non linearly-entangled predictive features diluted in a large number of irrelevant noisy variables, DL-based FS and SM interpretation methods are still far from being reliable.

A PSO-Optimized Neural Network and ABC Feature Selection Approach with eXplainable Artificial Intelligence (XAI) for Natural Disaster Prediction

Introduction: ”Artificial Intelligence will revolutionize our lives” is a phrase frequently echoed. The influence of Artificial Intelligence (AI) and Machine Learning (ML) extends across various aspects of our daily lives, encompassing health, education, economics, the environment, and more. Method: A particularly formidable challenge lies in decision support, especially in critical scenarios such as natural disaster management, where artificial intelligence significantly amplifies its ongoing capacity to assist in making optimal decisions. In the realm of disaster management, the primary focus often centers on preventing or mitigating the impact of disasters. Consequently, it becomes imperative to anticipate their occurrence in terms of both time and location, enabling the effective implementation of necessary strategies and measures. In our research, we propose a disaster forecasting framework based on a Multi-Layer Perceptron (MLP) empowered by the Particle Swarm Optimization (PSO) algorithm. The PSO-MLP is further fortified by the incorporation of the Artificial Bee Colony (ABC) algorithm for feature selection, pinpointing the most critical elements. Subsequently, we employ the LIME (Local Interpretable Model-agnostic Explanations) model, a component of eXplainable Artificial Intelligence (XAI). This comprehensive approach aims to assist managers and decision-makers in comprehending the factors influencing the determination of the occurrence of such disasters and increases the performance of the PSO-MLP model. The approach, specifically applied to predict snow avalanches, has yielded impressive results. Result: The obtained accuracy of 0.92 and an AUC of 0.94 demonstrate the effectiveness of the proposed framework. In comparison, the prediction precision achieved through an SVM is 0.75, while the RF classifier yields 0.86, and XGBoost reaches 0.77. Notably, the precision is further enhanced to 0.81 when utilizing XGBoost optimized by the grid-search. Conclusion: These results highlight the superior performance of the proposed methodology, showcasing its potential for accurate and reliable snow avalanche predictions compared to other established models.

Attention and dilated convolutions inclusive deep-CNN with multiplexed texture features to diagnose Pathological and High Myopia.

In today's era, precise and timely diagnosis of ocular diseases are crucial as these disorders jeopardize millions of visions. Early detection and proactive management can minimize vision threatening complications from these disorders. High Myopia(HM) and Pathological Myopia(PM), are the globally prevalent ocular diseases, which can impair vision acuity and productivity across all age groups. Routine screening via computer aided diagnosis(CAD) is required to detect HM and PM early and halt their progression. Previous studies majorly relied only on deep convolutional neural network(deep-CNN) features and independent analysis of PM and HM. This work seeks to provide a holistic analysis of PM and HM pathology by creating a novel ternary classifier on colour fundus images to detect normal vision, PM and HM respectively. We built the classifier by integrating texture features generated using gray level co-occurrence matrix(GLCM) within the deep-CNN model. Deep-CNN model comprises of spatial attention(SA) to locate lesion, squeeze-excitation(SE) to model interdependent channel attention & atrous or dilated convolutions to capture salient multiscale features of the related disease. Investigations employing extensive ablation techniques have elucidated the significance of optic disc(OD) and retinal vessel(RV) in the fundus images and their respective alterations in the HM or PM fundus. On the diverse dataset of 3212 colour fundus images, our ternary classifier has achieved 5-fold cross validation mean accuracy of 0.9754(+/-0.014), test accuracy of 0.9767 on 645 test fundus images and the kappa score of 0.9622 indicating the clinical viability of our classifier. The performance of our classifier excelled that of other pertinent studies for both PM and HM colour fundus images. Our classifier's test findings, comprehensive aetiology analysis and class activation maps are validated by the expert ophthalmologists making it reliable to serve as a virtual doctor alleviating the concerns with the lack of skilled ophthalmologists and expensive optometry tools by facilitating remote diagnostics and telemedicine applications.

Enhancing Sensitivity of Point-of-Care Thyroid Diagnosis via Computational Analysis of Lateral Flow Assay Images Using Novel Textural Features and Hybrid-AI Models.

Lateral flow assays are widely used in point-of-care diagnostics but face challenges in sensitivity and accuracy when detecting low analyte concentrations, such as thyroid-stimulating hormone biomarkers. This study aims to enhance assay performance by leveraging textural features and hybrid artificial intelligence models. A modified Gray-Level Co-occurrence Matrix, termed the Averaged Horizontal Multiple Offsets Gray-Level Co-occurrence Matrix, was utilised to compute the textural features of the biosensor assay images. Significant textural features were selected for further analysis. A deep learning Convolutional Neural Network model was employed to extract features from these textural features. Both traditional machine learning models and hybrid artificial intelligence models, which combine Convolutional Neural Network features with traditional algorithms, were used to categorise these textural features based on the thyroid-stimulating hormone concentration levels. The proposed method achieved accuracy levels exceeding 95%. This pioneering study highlights the utility of textural aspects of assay images for accurate predictive disease modelling, offering promising advancements in diagnostics and management within biomedical research.

Deep Learning for Detecting and Subtyping Renal Cell Carcinoma on Contrast-Enhanced CT Scans Using 2D Neural Network with Feature Consistency Techniques

Abstract Objective The aim of this study was to explore an innovative approach for developing deep learning (DL) algorithm for renal cell carcinoma (RCC) detection and subtyping on computed tomography (CT): clear cell RCC (ccRCC) versus non-ccRCC using two-dimensional (2D) neural network architecture and feature consistency modules. Materials and Methods This retrospective study included baseline CT scans from 196 histopathologically proven RCC patients: 143 ccRCCs and 53 non-ccRCCs. Manual tumor annotations were performed on axial slices of corticomedullary phase images, serving as ground truth. After image preprocessing, the dataset was divided into training, validation, and testing subsets. The study tested multiple 2D DL architectures, with the FocalNet-DINO demonstrating highest effectiveness in detecting and classifying RCC. The study further incorporated spatial and class consistency modules to enhance prediction accuracy. Models' performance was evaluated using free-response receiver operating characteristic curves, recall rates, specificity, accuracy, F1 scores, and area under the curve (AUC) scores. Results The FocalNet-DINO architecture achieved the highest recall rate of 0.823 at 0.025 false positives per image (FPI) for RCC detection. The integration of spatial and class consistency modules into the architecture led to 0.2% increase in recall rate at 0.025 FPI, along with improvements of 0.1% in both accuracy and AUC scores for RCC classification. These enhancements allowed detection of cancer in an additional 21 slices and reduced false positives in 126 slices. Conclusion This study demonstrates high performance for RCC detection and classification using DL algorithm leveraging 2D neural networks and spatial and class consistency modules, to offer a novel, computationally simpler, and accurate DL approach to RCC characterization.

Automatic Assessment of AK Stage Based on Dermatoscopic and HFUS Imaging-A Preliminary Study.

Background: Actinic keratoses (AK) usually occur on sun-exposed areas in elderly patients with Fitzpatrick I-II skin types. Dermatoscopy and ultrasonography are two non-invasive tools helpful in examining clinically suspicious lesions. This study presents the usefulness of image-processing algorithms in AK staging based on dermatoscopic and ultrasonographic images. Methods: In 54 patients treated at the Department of Dermatology of Poznan University of Medical Sciences, clinical, dermatoscopic, and ultrasound examinations were performed. The clinico-dermoscopic AK classification was based on three-point Zalaudek scale. The ultrasound images were recorded with DermaScan C, Cortex Technology device, 20 MHz. The dataset consisted of 162 image pairs. The developed algorithm includes automated segmentation of ultrasound data utilizing a CFPNet-M model followed by handcrafted feature extraction. The dermatoscopic image analysis includes both handcrafted and convolutional neural network features, which, combined with ultrasound descriptors, are used in support vector machine-based classification. The network models were trained on public datasets. The influence of each modality on the final classification was evaluated. Results: The most promising results were obtained for the dermatoscopic analysis with the use of neural network model (accuracy 81%) and its combination with ultrasound scans (accuracy 79%). Conclusions: The application of machine learning-based algorithms in dermatoscopic and ultrasound image analysis machine learning in the staging of AKs may be beneficial in clinical practice in terms of predicting the risk of progression. Further experiments are warranted, as incorporating more images is likely to improve classification accuracy of the system.

Extensions and Detailed Analysis of Synergy Between Traditional Classification and Classification Based on Negative Features in Deep Convolutional Neural Networks

In recent times, deep convolutional neural networks became an irreplaceable tool for pattern recognition in many different machine learning applications, especially in image classification. On the other hand, these models are often used in critical systems which are the reason for new and recent research regarding their robustness and reliability. One of the most important issues for these models is their susceptibility to different adversarial attacks. In our previous work Milošević and Racković (Neural Network World. 2019;29(4):221–34), and Milošević and Racković (Neural Comput Applic. 2021;33:7593–602), the new type of learning applicable to all the convolutional neural networks was introduced: the classification based on the negative features and the synergy of traditional and those newly introduced network models. In the case of partial inputs/image occlusion, it was shown that our new method creates models that are more robust and perform better when compared to traditional models of the same architecture. In this paper, some extensions of the earlier proposed synergy are given by introducing negatively trained features and additional synergy between four independent neural network models. A detailed analysis of the robustness of the newly proposed model is performed on EMNIST and CIFAR-10 image classification data sets in the case of the selected input occlusions and adversarial attacks. The newly proposed neural network architecture improves the robustness of the neural network and increases its resistance to various types of input damage and adversarial attacks.

A fisheye image correction method based on deep learning

Purpose Fisheye camera is a wide-angle camera with a large field of view, which can obtain more information than conventional cameras, but its own disadvantage of large distortion can lead to shape distortion of objects in the captured image. To correct lens distortion, most scholars use manual calibration methods. The traditional manual calibration method relies on complex data computation and specialized mathematical knowledge, but the method is complex and not universal. Considering the nonlinear features of neural networks can be used to fit the nonlinear distortion of distorted images, this paper aims to correct the distortion of a fisheye image based on convolutional neural networks. Design/methodology/approach By generating fisheye distortion images using readily available cityscape road data sets, this paper designs an effective correction model by taking advantage of the fact that the relationship between the pixel coordinates is basically stable after uniform distortion. In our training process, the corresponding distortion is firstly synthesized using the original image, the loss function is constructed using the geometric means of the Hough transform, the whole model is then trained with the help of the loss of the linear slope, and finally the predicted parameters are used to correct the fisheye image, which realizes the overall end-to-end framework. Findings Experimental results show that the method proposed in this paper outperforms similar methods in terms of correction performance. Originality/value This paper proposed a fisheye distortion correction algorithm based on convolutional neural network, which uses distortion-free images based on a generalized fisheye imaging model and maps to generate a fisheye distortion picture data set. It also builds a loss function training model based on line reconstruction error loss of Hough transforms.

Adaptive dimensional search-based orthogonal experimentation SSA (ADOX-SSA) for training RBF neural network and optimal feature selection

Adaptive dimensional search-based orthogonal experimentation SSA (ADOX-SSA) for training RBF neural network and optimal feature selection

A Novel and Powerful Dual-Stream Multi-Level Graph Convolution Network for Emotion Recognition.

Emotion recognition enables machines to more acutely perceive and understand users' emotional states, thereby offering more personalized and natural interactive experiences. Given the regularity of the responses of brain activity to human cognitive processes, we propose a powerful and novel dual-stream multi-level graph convolution network (DMGCN) with the ability to capture the hierarchies of connectivity between cerebral cortex neurons and improve computational efficiency. This consists of a hierarchical dynamic geometric interaction neural network (HDGIL) and multi-level feature fusion classifier (M2FC). First, the HDGIL diversifies representations by learning emotion-related representations in multi-level graphs. Subsequently, M2FC integrates advantages from methods for early and late feature fusion and enables the addition of more details to final representations from EEG samples. We conducted extensive experiments to validate the superiority of our model over numerous state-of-the-art (SOTA) baselines in terms of classification accuracy, the efficiency of graph embedding and information propagation, achieving accuracies of 98.73%, 95.97%, 72.74% and 94.89% for our model as well as increases of up to 0.59%, 0.32%, 2.24% and 3.17% over baselines on the DEAP-Arousal, DEAP-Valence, DEAP and SEED datasets, respectively. Additionally, these experiments demonstrated the effectiveness of each module for emotion recognition tasks.

Investigating a Toolchain from Trajectory Recording to Resimulation

The growing variety of transportation options and increasing traffic congestion pose new challenges for road safety. As a result, there is an intensified focus on developing automated driving features and assistance systems aimed at minimizing accidents caused by human errors. The creation of these systems requires a substantial amount of testing kilometers, with estimates suggesting that around 2.1 billion kilometers would be necessary to ensure that each situation pertinent to the driving function is encountered at least once with a probability of 50%. This paper advances the microscopic simulation of traffic scenarios beyond linear patterns, utilizing the open-source environment openPASS. It addresses the research question of whether existing microscopic simulations are able to realistically represent non-linear traffic scenarios. A comprehensive toolchain integrates simulation with video recordings and laser scans. The study compares recorded traffic flow data with simulations at a T-junction, assessing the realism of vehicle models and trajectory representation. Three scenarios are analyzed, considering vehicles and pedestrians. The 3D geometry of the scene was captured with a laser scanner, enabling the mapping of recorded video data onto a geo-referenced environment. Object trajectories were extracted using an ’Regions with Convolutional Neural Networks features’ object detector. While openPASS simulated vehicle and pedestrian behaviors effectively, limitations in trajectory variability and reaction times were observed. These findings highlight the need for more realistic behavior models. This research emphasizes the necessity for improvements to accommodate complex driving behaviors and pedestrian dynamics.

Fast prediction of key parameters in FEBA using the COSINE subchannel code and artificial neural network

Numerical techniques have emerged as an essential tool for operators and designers to preemptively acquire key parameters in accidents analysis. However, due to insufficient experience, it is difficult for them to obtain satisfactory numerical results. Moreover, the uncertainty analysis and quantification necessitate the simulation of a substantial number of samples, which requires a significant amount of computational time. Therefore, the development of a fast prediction model becomes imperative. In this work, a prediction model based on the in-house COSINE subchannel code and Multi-Head Perceptron (MHP) is developed. The COSINE subchannel code is employed to provide data sets for training neural networks. Firstly, the numerical results of COSINE subchannel code are compared with experimental data to ensure the accuracy of data sets. Secondly, input features for neural networks are selected by evaluating the impact of input parameters on numerical results, and a series of simulations is carried out to generate data sets. Then, a comparative analysis was conducted between the Multi-Layer Perceptron (MLP) and Support Vector Regression (SVR) models, and the MLP model performs better. Subsequently, the MLP was compared with the MHP, demonstrating the advantage of MHP model. Based on this, the predictions are conducted using the MHP model and the distribution of key parameters is compared with that obtained by COSINE subchannel code. The results illustrate that developed MHP model is an efficient tool for predicting key parameters during the reflooding phase.

Object Tracking Algorithm Based on Integrated Multi-Scale Templates Guided by Judgment Mechanism

The object tracking algorithm TransT, based on Transformer, achieves significant improvements in accuracy and success rate by fusing the extracted features of convolutional neural networks with the structure of Transformer. However, when dealing with the deformation of the object’s appearance, the algorithm exhibits issues such as insufficient tracking accuracy and drift, which directly affect the stability of the algorithm. In order to overcome this problem, this paper demonstrates how to expand scene information at the scale level during the fusion process and, on this basis, achieve accurate recognition and positioning. The predicted results are promptly fed back to the subsequent tracking process, from which temporal templates are embedded. Starting from both location and time can effectively improve the adaptive ability of the tracking model. In the final experimental comparison results, the algorithm proposed in this paper can adapt well to the situation of object deformation, and the overall performance of the tracking model has also been improved.

Network community detection via neural embeddings

Recent advances in machine learning research have produced powerful neural graph embedding methods, which learn useful, low-dimensional vector representations of network data. These neural methods for graph embedding excel in graph machine learning tasks and are now widely adopted. However, how and why these methods work—particularly how network structure gets encoded in the embedding—remain largely unexplained. Here, we show that node2vec—shallow, linear neural network—encodes communities into separable clusters better than random partitioning down to the information-theoretic detectability limit for the stochastic block models. We show that this is due to the equivalence between the embedding learned by node2vec and the spectral embedding via the eigenvectors of the symmetric normalized Laplacian matrix. Numerical simulations demonstrate that node2vec is capable of learning communities on sparse graphs generated by the stochastic blockmodel, as well as on sparse degree-heterogeneous networks. Our results highlight the features of graph neural networks that enable them to separate communities in the embedding space.

The unsupervised bearing fault diagnosis method based on the dual-framework Siamese network

Abstract Aiming at the problem of insufficient bearing fault samples in practical engineering, an unsupervised bearing fault diagnosis method with double frame twin network is proposed in this paper. This method only uses the source domain data to diagnose the target domain data, and can minimize the problem of domain difference. The concrete method is firstly to carry on continuous wavelet transform to the rolling bearing signal, and the processed signal is used as the input feature of convolutional neural network. The twin network structure is optimized and Manhattan distance is used as the loss function in the training process. The proposed CM network is applied to the diagnosis of inner ring fault, outer ring fault and normal state of rolling bearing. The results show that CM network can achieve over 90% accuracy on SQ variable speed vibration signal data set (SQV), Shijiazhuang Railway University data set (STD) and Case Western Reserve University data set (CWRU), which is suitable for bearing fault diagnosis under different working conditions and provides a new direction for fault diagnosis. The PyTorch code is available at https://github.com/xiaotianqu/The-unsupervised-bearing-fault-diagnosis-method-based-on-the-dual-framework-Siamese-network.

Real-time target recognition with all-optical neural networks for ghost imaging.

The generation and structural characteristics of random speckle patterns impact the implementation and imaging quality of computational ghost imaging. Their modulation is limited by traditional electronic hardware. We aim to address this limitation using the features of an all-optical neural network. This work proposes a real-time target recognition system based on an all-optical diffraction deep neural network for ghost imaging. We use a trained neural network to perform pure phase modulation on visible light, and directly complete the target recognition task by detecting the maximum value of light intensity signals at different positions. We optimized the system by simulating the effects of parameters, such as the number of layers of the network, photosensitive pixel, unit area etc., on the final recognition performance, and the accuracy of target recognition reached 91.73%. The trained neural network is materialised by 3D printing technology and experiments confirmed that the system successfully performs real-time target recognition at a low sampling rate of 1.25%. It also verified the feasibility and noise resistance of the system in practical application scenarios.

Deep Learning-Driven Sign Language Recognition: A Multimodal Approach for Gesture-to-Text Translation Using CNN-RNN Architectures

Sign language serves as a vital communication tool for individuals with hearing and speech impairments. However, a significant barrier exists when others do not possess a strong understanding of sign language often requiring interpreters to facilitate communication. To reduce the reliance on human interpreters, this research aims to develop an intelligent system capable of recognizing and translating sign language gestures into meaningful grammatically correct sentences. The proposed system processes both images and videos to interpret gestures, converting them into text that can be translated into any desired language. The system leverages state-of-the-art deep learning techniques, such as Convolutional Neural Networks (CNNs) for identifying key features in images and Recurrent Neural Networks (RNNs) for understanding temporal sequences in video data. By employing these advanced neural networks, the model is able to comprehend hand movements, facial expressions and other non-verbal cues to construct coherent sentences. Additionally, the system integrates natural language processing (NLP) to refine the output, ensuring the resulting sentences are grammatically correct. Our approach addresses common challenges such as differentiating between subtle hand gestures and reducing the impact of environmental noise in images. This solution holds the potential to significantly enhance communication for the hearing and speech impairedffering an efficient interpreter-free method of translating sign language into widely spoken languages

Development of an Artificial Intelligence System for Distinguishing Malignant from Benign Soft Tissue Tumors Using Contrast-Enhanced MR Images.

The integration of artificial intelligence (AI) into orthopedics has enhanced the diagnosis of various conditions; however, its use in diagnosing soft-tissue tumors remains limited owing to its complexity. This study aimed to develop and assess an AI-driven diagnostic support system for magnetic resonance imaging (MRI)-based soft-tissue tumor diagnosis, potentially improving accuracy and aiding radiologists and orthopedic surgeons. Experienced orthopedic oncologists and radiologists annotated 720 images from 77 cases (41 benign and 36 malignant soft-tissue tumors). Eleven tumor subtypes were identified and classified into benign and malignant groups based on histological diagnosis. Utilizing the standard machine learning classifier pipeline, we examined and down-selected imaging protocols and their predominant radiomic features within the tumor's three-dimensional region to differentiate between benign and malignant tumors. Among the scan protocols, contrast-enhanced T1 weighted fat-suppressed images showed the most accurate classification based on radiomics features. We focused on the two-dimensional features from the largest tumor boundary surface and its neighboring slices, leveraging texture-based radiomic and deep convolutional neural network features from a pretrained VGG19 model. The test data comprised 44 contrast-enhanced images (22 benign and 22 malignant soft-tissue tumors) containing six malignant and five benign subtypes distinct from the training data. We compared expert and non-expert human performances against AI by assessing malignancy detection and the time required for classification. The AI model showed comparable accuracy (AUC 0.91) to that of radiologists (AUC 0.83) and orthopedic surgeons (AUC 0.73). Notably, the AI model processed data approximately 400 times faster than its human counterparts, showcasing its capacity to significantly boost diagnostic efficiency. We developed an AI-driven diagnostic support system for MRI-based soft-tissue tumor diagnosis. While additional refinement is necessary for clinical applications, our system has exhibited promising potential in differentiating between benign and malignant soft-tissue tumors based on MRI.