Syed Adil Hussain Shah

Introduction: Breast cancer remains the most commonly diagnosed malignancy in women worldwide, with approximately 2.3 million new cases and 670,000 deaths in 2022. Artificial intelligence (AI) is increasingly being applied across breast imaging for detection, characterization, segmentation, risk stratification, explainability, and clinical translation. Methods: This systematic review synthesized 284 eligible peer-reviewed publications published between 2015 and 2025, identified through a PRISMA 2020-compliant search of four indexed databases. A supplementary relevance-ranked Google Scholar screen was reported separately. A targeted May 2026 narrative update of recent prospective and implementation studies was also conducted without adding these records to the PRISMA denominator. Results: The reviewed evidence encompassed classical machine learning and radiomics, convolutional neural networks, YOLO-family detectors, vision transformers and hybrid CNN-Transformer architectures, U-Net variants, Mask R-CNN, SAM, and MedSAM across mammography, tomosynthesis, MRI, ultrasound, contrast-enhanced mammography, and emerging photoacoustic imaging. The evidence highlights advances in explainable AI, multimodal and radiogenomic fusion, federated learning, and open-source deployment, while persistent challenges include data heterogeneity, class imbalance, domain shift, bias, incomplete calibration, and limited prospective validation. Retrospective benchmark findings were derived from the 284-study systematic corpus, whereas prospective clinical utility and real-world implementation were informed by the separate 2026 narrative update. Discussion: AI shows substantial potential to enhance breast cancer imaging, but broader clinical translation requires robust external and prospective validation, improved calibration, assessment of generalizability and bias, and integration into clinical workflows. Supplementary Data Sheet S1 should be interpreted as a provenance-tagged evidence map rather than a formal study-level risk-of-bias assessment, with 38 entries verified from full text, 48 based on abstracts, and 198 based on DOI/source pages or other web-accessible records.

Introduction: Voice-based digital biomarkers have emerged as a promising approach for distinguishing individuals with neurodegenerative disorders, particularly Parkinson’s disease (PD), from healthy subjects (HS). With the increasing availability of smartphone and web-based recording tools, voice data can be collected efficiently in both clinical and remote settings. However, further validation is required before such approaches can be translated into routine clinical practice. Methods: This study used a cross-sectional analysis at the recording level, treating repeated recordings from the same participant as separate observations collected at Vito Fazzi Hospital in Lecce, Italy. Speech recordings from individuals with Parkinson’s disease (PD) and healthy controls were collected using the dedicated Talia smartphone and web application. Sustained vowel phonation (/a/) was analyzed as the primary speech task. Following data acquisition, feature extraction was performed as a crucial step in the speech analysis pipeline, as the quality and relevance of the extracted features directly influence the ability of machine learning models to discriminate between Parkinson’s disease (PD) patients and healthy controls. To capture various aspects of speech impairment associated with PD, a comprehensive set of acoustic features was extracted, including long-term features (pitch, jitter, and shimmer), nonlinear descriptors such as Recurrence Period Density Entropy (RPDE), and short-term feature based on Mel-Frequency Cepstral Coefficients (MFCCs). These features were subsequently used to develop and evaluate machine learning models for the classification of Parkinson’s disease and healthy subjects. Feature selection was performed using SHAP to identify the most informative vocal biomarkers. Model performance was assessed using five independent random train–test splits (70% training and 30% testing), supported by an internal five-fold cross-validation procedure within the training data. Multiple machine learning models were developed and evaluated, including Random Forest, Logistic Regression, Support Vector Machine, Naive Bayes, K-Nearest Neighbors, Decision Tree, Artificial Neural Network, and Gradient Boosting. Results: The evaluated models demonstrated strong recording-level classification performance. Artificial Neural Networks (ANN) and K-Nearest Neighbors (KNN) achieved the highest accuracy scores (0.9545 and 0.9494, respectively), along with superior recall (up to 0.9500), precision (up to 0.9551), and F1-score (up to 0.9525). Both models also exhibited excellent discriminative ability, with ROC-AUC values reaching 0.9882 (ANN) and 0.9893 (KNN). In contrast, Naive Bayes and Decision Tree showed comparatively lower performance across all metrics. Log-loss analysis further confirmed the robustness of ANN and KNN, which achieved the lowest values (0.2552 and 0.2510, respectively), indicating well-calibrated predictions. Overall, the findings highlight the consistency and generalizability of ANN and KNN across cross-validation splits. Conclusions: This study demonstrates that machine learning models, particularly ANN and KNN, can effectively differentiate Parkinson’s disease from healthy conditions using voice recordings. The integration of explainable AI for feature selection enhances model transparency and clinical relevance. However, the reported performance estimates were obtained from a recording-level analysis and should be interpreted as preliminary findings. Further studies involving larger cohorts and participant-level validation strategies are required to determine the generalizability and clinical applicability of these approaches.

Preterm infants are at high risk of neurodevelopmental disorders (NDDs), yet current brain assessments using 2D cranial ultrasound (cUS) remain subjective and limited. 3D cUS offers richer anatomical details but is underutilized due to its complexity and lack of automated tools. We developed and evaluated Deep Learning (DL) methods for total brain (TB) segmentation from 3D cUS of preterm neonates, emphasizing accuracy, reproducibility, and clinical deployability. We compared 2D UNet models, a contextual UNet, and self-configuring 2D/3D nnUNet variants, evaluating accuracy, efficiency, and GPU/CPU feasibility. Clinical relevance was examined through longitudinal total brain volume (TBV) trajectories and their association with 2-year neurodevelopmental outcomes. The nnUNet ensemble achieved the best performance (Dice: 0.97, Volumetric Difference Error: 2.32%), while contextual UNet offered a favorable accuracy-efficiency trade-off on low-resource hardware. Longitudinal analysis showed significantly slower TBV growth in infants with adverse outcomes (p = 0.007), supporting the prognostic value of automated volumetric measurements. We developed a clinician-facing web application to illustrate clinical integration. This work demonstrates the feasibility of DL-based TB segmentation from 3D cUS and provides a scalable reproducible framework supporting early quantitative assessment of brain development, laying the groundwork for future lightweight, privacy-aware AI solutions for Neonatal Intensive Care Unit integration.

Accurate automated interpretation of upper-extremity musculoskeletal radiographs remains challenging because fracture appearance varies across anatomical regions and can be subtle under class imbalance. This study proposes a two-stage deep learning framework for MURA-based X-ray analysis, aiming to improve body-part recognition and body-part-wise abnormality detection. Multiple architectures were first compared for seven-class body-part classification, after which the selected hybrid Xception-Swin model was fine-tuned for abnormality detection within each anatomical subset. The framework combines Xception-derived local structural features with Swin Transformer contextual features using attention-based fusion, and performance was evaluated using accuracy, F1-score, AUC-ROC, Cohen’s kappa, calibration, component-level ablation, post hoc explainability, and zero-shot FracAtlas validation. For body-part classification, the model achieved accuracy = 0.9643, macro F1 = 0.9574, AUC-ROC = 0.9963, and kappa = 0.9579. For abnormality detection, accuracy ranged from 0.7289 to 0.8538, F1 from 0.7191 to 0.8508, AUC from 0.7693 to 0.9080, and kappa from 0.4449 to 0.7071. Ablation on hand and humerus radiographs showed the highest macro F1 with Hybrid Attention, while FracAtlas validation yielded AUC = 0.8247 and kappa = 0.5812. The results support complementary CNN-Transformer fusion and indicate preliminary cross-dataset generalizability. Implementation resources are available at Zenodo.

Plant diseases increasingly threaten global agriculture due to climate change, yet manual diagnosis remains challenging. We introduce B2-GraftingNet, a lightweight deep-learning framework for automated grape-leaf disease detection that combines a VGG16 backbone with Inception-style blocks to learn robust multi-scale cues. Binary Particle Swarm Optimization selects the most informative features before classification. On the public Kaggle grape-leaf dataset, a cubic SVM classifier achieves 99.56% peak accuracy, surpassing standard pretrained CNNs (VGG16/VGG19: 34.04%, Xception: 97.95%, Darknet: 94.91%, ResNet-50: 98.44%) while being faster and lighter. For transparency, we incorporate Grad-CAM, LIME, and occlusion-sensitivity with a local 20B-parameter LLM via Ollama, which processes ROI-aware XAI summaries and provides plain-language guidance in structured sections (verdict, where to look, what to do next, caution) without suggesting specific treatments.

Owing to changing climatic and environmental conditions, plant diseases are becoming increasingly prevalent, posing a serious threat to global agriculture. Timely and accurate diagnosis remains challenging, especially where scouting still relies on manual inspection. We propose B2-GraftingNet, a deep learning framework for automated detection of grape leaf diseases. B2-GraftingNet is a streamlined variant of our earlier B4-GraftingNet, retaining its strengths while simplifying blocks for faster inference and deployment. The architecture combines a VGG16 backbone with Inception-style blocks inside a custom CNN to extract robust, multi-scale features based on color, size, and shape. To reduce redundancy and improve generalization, Binary Particle Swarm Optimization (BPSO) selects informative features prior to classification. We evaluate Support Vector Machines (SVM) and k-Nearest Neighbors (KNN); a cubic SVM attains 99.56% peak accuracy on the public Kaggle grape-leaf dataset. For context, we also benchmarked standard pretrained CNNs on the same data, observing validation accuracies of 34.04% (VGG16), 34.04% (VGG19), 97.95% (Xception), 94.91% (Darknet), and 98.44% (ResNet-50); B2-GraftingNet matches or exceeds these while remaining lighter and faster to train and deploy. To enhance transparency and actionability, we pair Grad-CAM, LIME, and occlusion-sensitivity visualizations with a local gpt-oss:20b assistant (served via Ollama) that converts evidence into plain, grower-focused guidance and supports interactive chat validated by horticulturists. Results are further checked against expert-annotated ground-truth labels, confirming high accuracy and computational efficiency. Overall, B2-GraftingNet offers a reliable, interpretable, and scalable solution for early grape-leaf disease detection. The complete setup (code, model, web platform, configuration, and assets) is available on Zenodo: https://doi.org/10.5281/zenodo.17353656.

Speech Emotion Recognition (SER) is an umbrella term that encompasses all Machine Learning & Deep Learning (ML/DL) algorithms used for the very specific task of extracting emotional state from human speech. In literature, various techniques have been utilized to extract emotions from signals, including well-established speech analysis and classification techniques. Using the scoping review method, the paper maps techniques for speech-based emotion recognition. In doing so, it presents elements defining the use of algorithms to assess emotions, for example, databases used for emotion recognition, notions and types of emotions considered, and empirical investigations made toward SER and related limitations. The contribution places particular emphasis on the existing perspectives and practices in order to offer a series of recommendations for future developments.

The viral disease COVID-19, declared a pandemic by the World Health Organization (WHO), primarily affects the respiratory system and can be fatal. Although the Reverse Transcriptase Polymerase Chain Reaction (RT-PCR) test remains the gold standard for COVID-19 diagnosis, its time-intensive nature limits its effectiveness in urgent situations. To address this, we propose an ensemble of five state-of-the-art transfer learning (TL) models designed to mitigate biases and enhance the classification of COVID-19 from chest radiographs. A weighted optimization strategy combines the models, giving more weight to those with superior performance, ensuring more accurate and robust predictions. Evaluated on a publicly available dataset, the SqueezeNet model achieved the highest accuracy of 94.01% for three-class classification (Normal, COVID-19, Lung Opacity), while the ensemble approach achieved 92.57% accuracy and an F1-score of 92.36%, demonstrating resilience to transfer learning biases. This framework offers reliable diagnostic support, streamlining radiology workflows and enhancing decision-making in high-demand clinical environments. Additionally, it serves as a valuable tool for advancing medical artificial intelligence expertise among graduates.

Background: Traditional neonatal assessments rely on anthropometric measures such as weight, body size, and head circumference. However, recent studies suggest that objective movement quantification may serve as a complementary clinical indicator of development in preterm infants. Methods: This study evaluates non-invasive computer vision-based quantification of neonatal movement using contactless pose tracking based on computer vision. We analyzed approximately 800,000 postural data points from ten preterm infants to identify reliable algorithms, optimal recording duration, and whether whole-body or regional tracking is sufficient. Results: Our findings show that 30 s video segments are adequate for consistent motion quantification. Optical flow methods produced inconsistent results, while distance-based algorithms—particularly Chebyshev and Minkowski—offered greater stability, with coefficients of variation of 5.46% and 6.40% in whole-body analysis. Additionally, Minkowski and Mahalanobis metrics applied to the lower body yielded results similar to full-body tracking, with minimal differences of 0.89% and 1%. Conclusions: The results demonstrate that neonatal movement can be quantified objectively and without physical contact using computer vision techniques and reliable computational methods. This approach may serve as a complementary clinical indicator of neonatal progression, alongside conventional measures such as weight and size, with applications in continuous monitoring and early clinical decision-making for preterm infants.