Open research questions in Brain Tumor Detection and Classification

What the literature leaves open

• Future studies with a larger, more clinically representative negative cohort (including patients with hip pain but no labral injuries) are warranted to further validate the model’s generalizability and optimize its specificity for real world clinical practice.

  A YOLOv8 deep learning model for detecting labral injury via hip magnetic resonance imaging · 2026 · DOI

• Currently, brain tumor segmentation is confronted with challenges such as limited dataset sizes, imbalanced data low data collection quality. These categories, and challenges can directly or the indirectly performance of CNN in brain tumor segmentation. In the future, these issues might be resolved by adopting more efficient data augmentation techniques, devising more training methods, and exploring effective model unsupervised learning approaches.

  Retraction Notice: Improvements in Brain Tumor Segmentation Methods Based on Convolutional Neural Networks · 2026 · DOI

• Although the proposed COA–SVD–U-Net framework shows promising results for automated brain tumor seg- mentation from MRI images, several improvements can be Page 11 of 12 496 explored in future research. First, the model should be eval- uated on larger and more diverse datasets collected from multiple hospitals to improve its generalization capability in real clinical environments. Testing the framework on multi- institutional MRI datasets will help ensure robustness under different imaging conditions and patient variations. Second, future work can integrate the proposed optimi- zation strategy with more advanced deep learning architec- tures such as TransUNet and UNETR, which have recently shown strong performance in medical image analysis. Such integration may improve feature learning and segmentation accuracy. In addition, extending the current two-dimen- sional approach to volumetric segmentation using 3D U-Net could help capture three-dimensional tumor structures more effectively. Future studies may also explore multimodal MRI data such as T1, T1c, T2, and FLAIR images to obtain richer information about tumor tissues. incorporating explain- able AI techniques can improve model interpretability and increase trust among clinicians. Finally, optimizing the framework for real-time clinical deployment and validat- ing it with medical experts will help transform the proposed method into a practical computer-aided diagnosis system for brain tumor detection. Author Contributions All authors contributed significantly to this re- search work. Conceptualization and methodology: Ajay Singh Yadav, Navin Ahlawat. Mathematical modeling and algorithm design: Ajay Singh Yadav, Priyanka Agarwal. Software implementation and experi- ments: Anjali Malik, Anupam Swami, M. Vinothkumar. Data analysis and validation: Bhavani Viswanathan, Garima Pandey. Writing—orig- inal draft preparation: Ajay Singh Yadav, Anjali Malik. Writing—re- view and editing: All authors. Supervision: Ajay Singh Yadav. Funding The authors declare that no funding was received for this research work. This study was carried out as part of the authors’ aca- demic and institutional research activities without any financial sup- port from external funding agencies. Data Availability The datasets used in this study are publicly available. The primary dataset utilized is the BRATS (Brain Tumor Segmenta- tion) dataset, which can be accessed through official repositories. Additional processed data and implementation details supporting the findings of this study are available from the corresponding author upon reasonable request.

  Integrated Brain Tumour Detection: Cuckoo Optimization and SVD- Guided Segmentation with U-Net · 2026 · DOI

• This study addresses the difficulty of detecting BT in highly imbalanced MRI datasets by offering an integrated system that combines ConvNeXtTiny (feature extraction), SMOTE (class balancing), ETC (feature selection), and stacking. The ConvNeXtTiny accurately captured small MRI features, whereas SMOTE assured equitable represen- tation of minority classes. ETC selected the most relevant 2,000 features and stacked them to enhance model resilience. The ConvNeXtTiny-SMOTE-ETC-SM obtained 100.0% accuracy, F1-score, MCC, and kappa on the dataset, and 95.8% accuracy, 95.9% F1, 92.0% MCC, 91.7% kappa, and p_value = 0.005 after 10-fold cross-validation. Confi- dence intervals calculated using NAI and bootstrap verification confirmed the model’s reliability (LCI: 89.5–94.0%, UCI: 100%). LIME-based interpretability improved confi- dence and transparency. A comparative examination revealed that the proposed tech- nique outperformed four cutting-edge TL models (VGG16, InceptionV3, MobileNetV2, and Xception), demonstrating its superiority. Despite promising outcomes, a few caveats must be recognized: • Data limitations: The dataset had a class imbalance and lacked diversity, which could limit its applicability to larger patient populations. • Absence of external validation: The approach was not evaluated on separate, multi- institutional information sets, which is required to ensure robustness and practical dependability. • Computational complexity: ConvNeXtTiny and stacking require more training time • and computational cost, which may limit adoption on low-resource platforms. Interpretability challenges: Despite the use of LIME, interpretability remains limited, and physicians may require more comprehensible explanations of the model’s findings. Future studies will concentrate on the following areas to improve clinical relevance and generalization: • • • • • • Independent and large-scale verification: Test the model on bigger, larger, and real- world medical MRI datasets to guarantee its resilience across demographics and scanners. Multi-modal fusion: Utilize alternative data sources, such as genomes, clinical information, or 3D MRI sequences, to enhance diagnostic accuracy. Adaptable assistance with decision tools: Create intuitive interfaces with explainable AI elements to help clinicians make real-time decisions. Enhanced Interpretability Techniques: Interpretability of artificial intelligence [59], Utilize SHAP, Grad-CAM, or concept-based presentations to increase confidence and transparency. Consider transformer-based architectures [60], quantization, or light designs to decrease processing and facilitate real-time distribution. Transformer-based models [60, 61] and Generative Adversarial Networks (GAN) [62, 63] can be used for other brain illnesses, such as brain deformation [64] and Lanjewar et al. Discover Computing (2026) 29:342 Page 24 of 26 malignancies [65]. Furthermore, non-destructive techniques like near-infrared spectroscopy [66], Magnetic particle imaging (MPI) [67], and Non-invasive electroporation method like pulsed magnetic field (PMF) [68] can be combined with conventional diagnostic approaches to improve performance.

  Enhanced brain tumour detection using fusion of ConvNeXtTiny with stacking generalization · 2026 · DOI

• regarding for the best choices automatic discovery of features and end-to-end learning process. Among deep learning methods, U-Net became one of the most efficient tools for biomedical image segmentation due to the implementation of an encoder-decoder architecture and skip connections that allow retaining spatial features. On the other hand, image classification models such as EfficientNet family of architectures perform better by employing compound scaling strategies and simultaneously optimizing network depth, width, and resolution. In this regard, EfficientNet-B7 is considered one of image classification tasks. However, some recent progress has been made in the context of transformer architectures and Vision Mamba, allowing longer dependencies within medical images. Despite the advantages gained by the new technologies, transformer models still require huge computational resources and large training datasets. Also, many existing methods focus only on one it segmentation or task, be classification, but not both. Another important challenge that exists when designing medical AI systems is that of interpretability. Deep learning models can often be viewed as "black boxes," meaning that physicians find it difficult to determine how their predictions. In particular, Explainable Artificial Intelligence (XAI), including Grad-CAM, provides visualized interpretations through showing parts of input data that help form decisions.

  A Robust MRI-Based Brain Tumor Diagnosis Framework Integrating U-Net, EfficientNet-B7, and Grad-CAM Visualization · 2026 · DOI

• 9.1 CONCLUSION The Brain Tumor Classification System presented in this project demonstrates the effective use of deep learning and transfer learning techniques for medical image analysis. By leveraging EfficientNetB0 as a pre-trained backbone, the system is capable of accurately classifying MRI brain scans into four categories: glioma, meningioma, pituitary tumor, and no tumor. The approach combines robust feature extraction, fine-tuning, and data augmentation to achieve high accuracy while maintaining generalizability across previously unseen MRI images. The inclusion of histogram equalization as an enhancement technique provides better visual clarity for users without altering the input used for prediction, thus separating visualization from model training. The system’s three-column graphical user interface (GUI) enhances usability, allowing users to simultaneously view the original image, the enhanced version, and the classification results along with class probabilities. This interface ensures that even non-technical users, such as clinicians and students, can interact with the system efficiently. The design choices, including status indicators, probability displays, and interactive buttons, provide a seamless workflow from image upload to result interpretation. From a research perspective, the project validates the effectiveness of transfer learning for medical imaging, particularly in cases where labeled datasets are limited. The two-stage training process—first training only the top layers and then fine-tuning the deeper layers of EfficientNetB0—demonstrates how pretrained networks can be adapted to specific tasks with minimal risk of overfitting. Additionally, the use of modern image preprocessing and augmentation techniques enhances model robustness, making it more resilient to variations in MRI scans, such as differences in scanner settings, slice orientation, and patient anatomy. In terms of clinical relevance, this system has the potential to assist radiologists in early tumor detection, provide second-opinion support, and accelerate the diagnostic workflow. While the current model focuses on classification, the framework is extendable for tumor segmentation, multi-modal MRI analysis, and integration into hospital PACS systems. Future improvements, such as incorporating explainable AI techniques like Grad- CAM, will allow clinicians to visualize and validate model decisions, thereby increasing trust and adoption in real-world medical settings. Overall, the Brain Tumor Classification System represents a significant step toward AI-assisted diagnostic tools in neuro-oncology. It combines the strengths of modern deep learning architectures with a user-friendly interface, providing an effective, accessible, and accurate solution for brain tumor classification. Continuous dataset expansion, model refinement, and feature enhancements will further improve its performance and clinical utility, paving the way for a practical, deployable tool in medical imaging workflows. © 2026 The Author(s). Published by IJCOPE Journal. Website: https://ijcope.org/ 29 International Journal of Creative and Open Research in Engineering and Management ISSN: 3108-1754 (Online) Volume 02 Issue 04 April-2026 | Impact Factor: 3.5 FUTURE ENHANCEMENTS  Expand Dataset: Incorporate more diverse and larger datasets to improve model robustness and reduce class imbalance.  Multi-modal MRI Input: Include different MRI modalities (T1, T2, FLAIR) for more accurate tumor detection.  Explainable AI: Integrate Grad-CAM or other visualization methods to highlight tumor regions and improve interpretability.  Edge Deployment: Optimize the model for deployment on mobile or edge devices for real-time diagnosis.  Segmentation Module: Extend the system to not only classify but also segment tumor regions for surgical planning.  Continuous Learning: Implement online learning to update the model with new patient data over time.  Integration with PACS: Enable automatic fetching and classification of MRI scans directly from hospital PACS systems.  Enhanced GUI Features: Add batch processing, 3D visualization of MRI slices, and reporting tools for clinicians.

  AI -Based Novel Cognitive Computational Strategy for Optimizing Brain Tumor Classification Using Magnetic Resonance Imaging Data · 2026 · DOI

• increasingly high, methodological and reporting currently hinder into routine clinical practice. Addressing these gaps—particularly in dataset diversity, protocol transparency, and validation strategy—represents a necessary step toward reliable clinical deployment. Collectively, these challenges emphasize the need for larger, diverse, and well-annotated datasets; standardized reporting of PG & PA features; inclusion of multimodal clinical data; and external validation. As segmentation models advance, these methodological improvements will be critical to ensuring their reliability, generalizability, and eventual clinical adoption. 4.2.1 Study limitations In this systematic review, a noteworthy limitation is the time interval between the last database search (October 2024) and the completion of this manuscript (July 2025). Given the rapid pace of developments in medical image segmentation and deep learning, it is possible that additional studies have since been published that further advance the field. In addition, a quantitative meta-analysis was not feasible due to the heterogeneity and limitations in the reporting of MRI acquisition parameters, dataset characteristics, and PG & PA features. 4.3 Future directions Future work should focus on developing generalisable models for PG and PA segmentation across diverse imaging protocols and populations, supported by large, multi-institutional annotated datasets. Integrating multimodal information—such as clinical or endocrine markers—may enhance diagnostic relevance and bridge the gap between imaging and functional assessment. Standardised benchmarking frameworks and rigorous external fair comparison across to enable validation are essential methods and to support clinical translation. From a clinical implementation perspective, future studies should be designed as prospective or retrospective clinical validation studies that explicitly account for the methodological limitations identified in this review, including variability in imaging protocols, limited reporting of acquisition parameters, and tumour characteristics. Addressing these factors within study design would facilitate more robust evaluation of model reliability and generalisability.

  Systematic review of pituitary gland and pituitary adenoma automatic segmentation techniques in magnetic resonance imaging · 2026 · DOI

• The quantitative evaluation uses eight mathematical metrics (EN, MI, SD, SF, PSNR, AG, CC, SCD) but provides limited justification for why these specific metrics were selected or how they correlate with actual clinical diagnostic utility in detecting metabolic anomalies in brain tissue.

  Bridging Modalities through Deep Networks: Advanced Medical Fusion Image Techniques · 2026 · DOI

• The paper does not address how the Med Fuse architecture generalizes to other multi-modal medical imaging combinations (e.g., CT-SPECT, fMRI-structural MRI, ultrasound-MRI) or whether retraining is required for different modality pairs beyond PET-MRI fusion.

  Bridging Modalities through Deep Networks: Advanced Medical Fusion Image Techniques · 2026 · DOI

• The clinical validation of Med Fuse is absent; no radiologist reader studies or inter-observer agreement assessments comparing diagnostic confidence and accuracy between Med Fuse-generated fusions and conventional mathematical fusion methods are reported.

  Bridging Modalities through Deep Networks: Advanced Medical Fusion Image Techniques · 2026 · DOI

• While the paper demonstrates successful rejection of non-medical inputs through heuristic validation, no ablation study is provided to assess which specific architectural components (e.g., colormap application, Min-Max normalization, deterministic reporting engine) contribute most to preventing color bleeding and artifact generation in the CNN.

  Bridging Modalities through Deep Networks: Advanced Medical Fusion Image Techniques · 2026 · DOI

• The paper compares Med Fuse against five conventional mathematical fusion methods (PCA, DWT, CVT, LatLRR, DenseFuse) but does not benchmark against other deep learning-based fusion approaches or transformer-based architectures. Direct comparison with recent neural network fusion methods is absent.

  Bridging Modalities through Deep Networks: Advanced Medical Fusion Image Techniques · 2026 · DOI

• The Med Fuse CNN was evaluated on only 20 MRI-PET image pairs; the dataset size is insufficient to establish generalizability across diverse patient populations, imaging protocols, and scanner manufacturers. Validation on larger, multi-center datasets with varying acquisition parameters is needed to confirm the robustness of the deep learning fusion architecture.

  Bridging Modalities through Deep Networks: Advanced Medical Fusion Image Techniques · 2026 · DOI

• The proposed Dual-3DM3-AD model for Alzheimer's diagnosis presents its practical implementation in real clinical environments. Firstly, the model's reliance on high-quality and diverse MRI and PET datasets may pose challenges in real-world settings, where data availability can be limited. Additionally, the computational demands of the model, including network architectures, may strain the resources of healthcare facilities. The lack of model interpretability hinders the understanding of how diagnoses are arrived at, potentially impacting trust among healthcare professionals. Variations in imaging standards and equipment in clinical settings must be addressed for the model to perform consistently. preprocessing complex neural and VI. CONCLUSION AND FUTURE WORK Lack of training/testing data consideration and ineffective segmentation are one of the major reasons for low Alzheimer diagnosis accuracy, which is still a crucial concern. To alleviate these issues, we presented a promising avenue for a more comprehensive understanding of AD staging. This paper introduced an innovative approach to address this challenge. We proposed the Dual-3DM3-AD model, designed for accurate and early Alzheimer's diagnosis, by leveraging both MRI and PET image scans. Our methodology involved a series of preprocessing steps, including noise reduction, skull stripping, and 3D image conversion, performed using the QNLM, Morphology function, and BDM, respectively, to enhance the image quality. Subsequently, we employed a Mixed-transformer with Furthered U-Net architecture for semantic segmentation, effectively reducing complexity. The Dual-3DM3-AD model incorporated a multi-scale feature extraction module to extract pertinent features from the segmented images. These extracted features were then aggregated using the densely connected feature aggregator module to make the most of both information sources. Furthermore, we employ a multi-head attention mechanism to reduce feature dimensionality, followed by the application of a softmax layer for multi-class Alzheimer's diagnosis. Our proposed Dual-3DM3-AD model was implemented in MATLAB 2020A and rigorously compared with several baseline approaches by using a range of performance metrics, including accuracy, sensitivity, specificity, f-measure, and ROC curve analysis. Remarkably, our work surpassed existing models in multi-class Alzheimer's diagnosis, underscoring its potential as a valuable tool in the early detection of this debilitating disease. In terms of future work, we have planned to propose an Explainable Artificial Intelligence (EAI) with computation for better understanding of classification result with the aim of further reducing computational complexity and including feedback system.

  Dual-3DM³AD: Mixed Transformer Based Semantic Segmentation and Triplet Pre-Processing for Early Multi-Class Alzheimer’s Diagnosis · 2024 · DOI

• However, it was stated that CNN was insufficient to measure the relationship between these features while extracting image features, and it could not hide features such as pose (position, direction, size), deformation, and texture.

  Performance Evaluation of Basic Capsule Network Architecture in Classification of Biomedical Images · 2023 · DOI

• Resampling and cost-sensitive learning remain established practices for skewed label distributions, often paired with trans- fer learning to leverage pre-trained representations when data is scarce.

  A comprehensive review of deep learning algorithms for imbalance medical image processing: recent developments and future opportunities · 2026 · DOI

• Existing studies have shown that watermarking has a negligible effect on medical image classification; nevertheless, its impact on segmentation performance remains insufficiently explored.

  Nested Attention Network for Robust Medical Image Segmentation Under Digital Watermarking · 2026 · DOI

• Because performance alone is insufficient, interpretability has become central, with saliency maps and visual explanations increasing trust, and fairness audits identifying subgroup differences that can otherwise undermine equity.

  Systematic Review for Detecting & Diagnosing Brain Tumor Using Deep Learning: Current Trends and Future Research Directions · 2026 · DOI

• Reliable classification of brain tumors remains challenging for computer-aided diagnosis, since each tumor type can look very different, only sparse data exist, and images and genomic profiles contain noise.

  A Global-Local Interaction Modeling Network with Adaptive Feature Optimization for Brain Tumor Classification Using MRI · 2026 · DOI

Most-cited papers in Brain Tumor Detection and Classification

• Employing deep learning and transfer learning for accurate brain tumor detection · Scientific Reports · 2024 · 218 citations
• MedSegDiff-V2: Diffusion-Based Medical Image Segmentation with Transformer · Proceedings of the AAAI Conference on Artificial Intelligence · 2024 · 211 citations
• Brain tumor segmentation in MRI with multi-modality spatial information enhancement and boundary shape correction · Pattern Recognition · 2024 · 177 citations
• Automated Brain Tumor Segmentation and Classification in MRI Using YOLO-Based Deep Learning · IEEE Access · 2024 · 171 citations
• Brain tumor detection from images and comparison with transfer learning methods and 3-layer CNN · Scientific Reports · 2024 · 162 citations
• A novel CNN architecture for accurate early detection and classification of Alzheimer’s disease using MRI data · Scientific Reports · 2024 · 159 citations
• MRI brain tumor detection using deep learning and machine learning approaches · Measurement Sensors · 2024 · 158 citations
• Optimized Brain Tumor Detection: A Dual-Module Approach for MRI Image Enhancement and Tumor Classification · IEEE Access · 2024 · 157 citations
• A hybrid deep CNN model for brain tumor image multi-classification · BMC Medical Imaging · 2024 · 151 citations
• Enhancing brain tumor detection in MRI images through explainable AI using Grad-CAM with Resnet 50 · BMC Medical Imaging · 2024 · 146 citations

Most recent work

• Enhanced ResU-Net for brain tumor segmentation using EfficientNetB0, Channel Attention, and ASPP · Biomedical Physics & Engineering Express · 2026
• Comparative evaluation and interpretability analysis of modern CNN architectures for brain tumor MRI classification · Journal of Associated Medical Sciences · 2026
• Development of Brain MRI Image Segmentation program using UNET++ network for radiosurgery planning · Nuclear Science and Technology · 2026
• A Deep Learning Framework for Brain Tumor Classification Using VGG16-Based Autoencoder with BiLSTM Feature Refinement · International Journal on Smart Sensing and Intelligent Systems · 2026
• Collective Intelligence: Harnessing Multiple Models to Enhance Brain Tumor Detection and Classification Utilizing Deep Learning Techniques · Cureus Journal of Computer Science · 2026
• Systematic Review for Detecting & Diagnosing Brain Tumor Using Deep Learning: Current Trends and Future Research Directions · International Journal of Computational Intelligence Systems · 2026
• Privacy-Preserving Brain Tumor Classification using VIT, Eigencam and Federated Learning · International Journal of Science, Strategic Management and Technology · 2026
• Bridging Modalities through Deep Networks: Advanced Medical Fusion Image Techniques · International Journal for Research in Applied Science and Engineering Technology · 2026
• Deep Learning-Based Cancer Diagnosis Using a Bluefin Trevally-Optimized Convolutional Neural Network · Engineering Technology & Applied Science Research · 2026
• Entropy-Based Convolutional Layer Selection for Transfer Learning in Brain Tumor Detection · Arabian Journal for Science and Engineering · 2026

Related topics in Neuroscience