Shahid Mohammad Ganie

Background: Panoramic dental radiographs frequently contain multiple coexisting pathological, anatomical, and treatment-related findings that vary in scale, contrast, and boundary definition. Although recent one-stage detectors provide strong performance, reliable localization of small or weakly defined findings remains challenging. Methods: We propose SBRF-Det, a one-stage multi-finding detector integrating scale-dependent representative allocation with boundary-conditioned local–global feature fusion at P4/16. The method was evaluated on OralXrays-9, comprising 12,688 panoramic radiographs and 84,099 retained annotations across nine categories, using a fixed 2688-image hold-out cohort. Twenty-two detector configurations were compared under a common protocol, and A0–A2 were further assessed using three independent seeds and paired radiograph-level bootstrap analysis. Results: SBRF-Det achieved 0.8777 precision, 0.8979 recall, 0.9152 mAP50, and 0.7029 mAP50-95. Relative to the matched YOLO12l baseline, mAP50-95 increased by 0.95 percentage points. Performance remained class dependent, with apical periodontitis reaching 0.4526 mAP50-95 versus 0.8401–0.8842 for implants, porcelain crowns, and ceramic bridges. SBRF-Det used 93.06 GFLOPs versus 95.99 for A0, but inference latency increased from 5.946 to 8.851 ms/image. Three-seed experiments preserved the A0–A1–A2 ordering, while paired bootstrap intervals for A2–A0 and A2–A1 excluded zero. Conclusions: SBRF-Det improves strict localization through representative and boundary-aware feature modeling, but the gain is modest and introduces a runtime trade-off. Patient-linked, multi-center, and reader-centered validation remains necessary before clinical deployment.

Lung cancer remains one of the leading causes of cancer-related mortality worldwide, where early detection and reliable risk assessment are critical for improving outcomes. This study develops and evaluates a range of ensemble learning models for lung cancer prediction using lifestyle and clinical indicators, with an emphasis on both predictive performance and interpretability. Five base learners—logistic regression, k-nearest neighbors, naïve Bayes, support vector machine, and linear discriminant analysis—were used to construct multiple boosting and bagging models. Building on these, voting and stacking ensembles were designed by selectively combining high-performing models. All approaches were evaluated on the original dataset as well as on balanced and upsampled variants derived through synthetic augmentation. Model performance was assessed using accuracy, precision, recall, F1-score, Matthews correlation coefficient (MCC), and AUC-ROC. The results show that ensemble approaches consistently outperform individual models, with voting and stacking demonstrating superior performance over boosting and bagging methods. The stacking model achieved the strongest overall performance across all evaluated models. On the original dataset, which provides a more realistic representation of practical deployment conditions, it attained an accuracy of 93.53%. Performance further improved on the balanced and upsampled datasets on the upsampled dataset. To enhance transparency, SHAP and LIME were employed to provide global and local interpretability, respectively, enabling identification of key clinical factors and patient-specific risk drivers. The analysis highlights both alignment with known clinical patterns and dataset-driven variations, supporting informed interpretation of model outputs. The results suggest that stacking-based ensembles can improve risk prediction from lifestyle and clinical indicators while maintaining model transparency through SHAP and LIME explanations. These findings highlight the potential of interpretable ensemble learning as a decision-support tool for lung cancer risk assessment.

Introduction: Accurate detection of gastrointestinal diseases from endoscopic images remains challenging due to substantial inter-class similarity, intra-class variability, and heterogeneous lesion presentation. While convolutional neural networks (CNNs) effectively capture fine mucosal textures, their limited receptive field restricts global contextual reasoning. Conversely, transformer architectures model long-range dependencies but may underrepresent localized structural detail. Methods: This study proposes a dual-branch hybrid transformer framework that integrates global token-level reasoning from ViT-L/32 with hierarchical spatial modeling from MaxViT-L. An adaptive gated fusion mechanism is introduced to dynamically regulate the relative contribution of local and global representations on a per-sample basis, enabling content-aware cross-scale integration. The model was evaluated on a multi-class gastrointestinal endoscopic dataset using comprehensive quantitative metrics, class-wise analysis, ablation studies, confidence interval estimation, and statistical validation. Post-hoc explainability was further investigated using multiple complementary XAI techniques. Results: The proposed framework demonstrated consistently high accuracy, sensitivity, specificity, and area under the ROC curve across eight gastrointestinal disease categories, outperforming several state-of-the-art transformer and hybrid architectures. The XAI analyses further confirmed lesion-focused activation consistency and stable decision-making. Discussion: The findings indicate that adaptive local–global feature integration through gated fusion enhances both diagnostic performance and interpretability, supporting the potential of hybrid transformer architectures for reliable computer-aided gastrointestinal disease classification.

Background: Accurate classification of brain tumors in MRI scans is critical for effective clinical decision making, yet manual assessment is labor-intensive and subject to variability. Although deep learning models, particularly CNNs and Transformers, have shown potential, each architecture alone has limitations: CNNs excel at local feature extraction but lack global context awareness, while Transformers capture global relationships but may overlook fine-grained details. Objectives: To develop a hybrid Transfer Learning–Transformer model that integrates CNN-driven local feature modeling with Transformer-based global reasoning to enhance multi-class brain tumor classification. Methods: The proposed workflow comprises three stages: (1) benchmarking standalone CNN (ResNet50, VGG19, ConvNeXtBase, EfficientNetV2B0) and Transformer models (ViT, Swin, DeiT, PoolFormer); (2) constructing two intra-family hybrids—H _TL (ResNet50 + ConvNeXtBase) and H _TF (PoolFormer + ViT); and (3) fusing them into a final hybrid model (H _F ). Two public MRI datasets (Figshare, Kaggle) were used. Models were trained, validated, and tested on the Kaggle dataset, while the Figshare dataset was used exclusively for external validation. Performance was assessed using accuracy, precision, recall, F1-score, AUC, Friedman’s aligned-rank test, Kendall’s W, Holm post hoc analysis, TOPSIS-based ranking, calibration analysis (Brier score), and Grad-CAM++ for interpretability. Results: The H _F model achieved near-perfect classification (∼99.4% on Kaggle and up to 100% on Figshare), significantly outperforming all baseline and single-architecture models. Statistical and multi-criteria analyses consistently ranked H _F as the top-performing model, while calibration results confirmed reliable probability estimation. Grad-CAM++ further indicated tumor-focused decision-making. Conclusion: The proposed hybrid model delivers high accuracy, strong generalizability, and reliable, interpretable predictions across MRI datasets, positioning it as a promising solution for AI-assisted brain tumor diagnosis.

Rapid and reproducible interpretation of electrocardiograms (ECGs) is essential for distinguishing acute myocardial infarction from chronic infarction patterns, rhythm abnormalities, and normal recordings. This study presents a controlled comparison of three transfer-learned Vision Transformer architectures—DeiT-Small, Swin-Tiny, and ViT-Tiny—for four-class classification of 12-lead ECG report images. A total of 3,951 images were stratified into training (n = 3, 159), validation (n = 393), and independent test (n = 399) subsets representing abnormal heartbeat, history of myocardial infarction, myocardial infarction, and normal ECG. All models were initialized with ImageNet-pretrained weights and fine-tuned under the same input resolution, augmentation policy, optimizer, and stopping criterion. Performance was assessed using accuracy, macro-precision, macro-recall, macro-F1 score, parameter count, floating-point operations, and inference latency. DeiT-Small achieved the highest macro-F1 score (0.9971) with an accuracy of 0.9975, while ViT-Tiny reached the same accuracy and a macro-F1 score of 0.9969 using only 5.53 million parameters, 2.1493 GFLOPs, and 0.1275 ms per image. Swin-Tiny remained accurate (0.9825) but required the greatest computational cost. The DeiT-Small confusion matrix contained a single error, in which one normal ECG was assigned to the history-of-myocardial-infarction class. These findings indicate that compact global-attention models can retain near-perfect internal discrimination while substantially reducing computational demand. Nevertheless, patient-independent and multi-center external validation is required before clinical use.

Background/Objectives: Alzheimer’s disease (AD), the most prevalent form of dementia, is characterized by an overall decline in cognitive functioning and represents a major public health crisis. It remains critical to be able to accurately and quickly diagnose patients with AD; however, recent deep learning approaches using MRI data do not provide sample generalization, have high computational requirements, and offer little interpretability. Methods: In this study, we present a new framework called eXplorative ViT-CNN (X-ViTCNN) that combines a customized Vision Transformer model with two previously trained CNNs (DenseNet201 and MobileNetV2). With our proposed preprocessing approach using contrast-enhanced preprocessing to highlight neuroanatomical features as well as Bayesian Optimization to tune hyperparameters, we fuse local structural features originating from the CNNs with global representations from the transformer and feed the final result to fully connected dense layers for multi-stage classification. We also use Grad-CAM visualizations to provide insight into how our model arrived at its classification. Results: Experiments conducted on ADNI and OASIS datasets demonstrate the superiority of X-ViTCNN, achieving accuracies of 97.98% and 94.52%, respectively. The model outperformed individual baselines and other pre-trained architectures, showing balanced sensitivity and specificity across all AD stages. Conclusions: The proposed X-ViTCNN framework is a powerful, interpretable method for predicting the development of multi-stage Alzheimer’s disease using MRI scans. The combination of complementary feature learning, automatic hyperparameter optimization and interpretability through visualization make it an excellent potential tool for clinicians to support their decision making in the early diagnosis and ongoing monitoring of persons with Alzheimer’s disease.