Dr. Shrooq Alsenan is an Assistant Professor and the Director of CCIS AI Center at Princess Nourah bint Abdulrahman University. Dr. Shrooq is holding a PhD in Information Systems Sciences from King Saud University specializing on Artificial Intelligence applications in healthcare, medicine and Drug discovery. Her research was awarded "1st place distinguished PhD dissertation" in the Annual Awards Ceremony for Excellence in Scientific Research at King Saud University 2022. She was also recipient of the healthcare innovation research chair grant at KSU.
Dr. Shrooq received two fellowship grants from IBK Program for Saudi women and from MIT Jameel Clinic enabling her to assume the role of a Postdoctoral Massachusetts Institute of Technology (MIT). She worked in Computer Science & Artificial Intelligence Lab (CSAIL) and Jameel Clinic AI & Healthcare Center at MIT.
EDUCATION
PhD in Information Systems Sciences from King Saud University specializing on Artificial Intelligence applications in healthcare
RESEARCH, TEACHING, or OTHER INTERESTS
Artificial Intelligence, Information Systems
57
Scopus Publications
Scopus Publications
A gated task-attentive multi-task network for unified retinal image analysis Muhammad Zaheer Sajid, Imran Qureshi, Muhammad Fareed Hamid, Mohammad Alhefdi, Shrooq Alsenan, Qaisar Abbas, Yongwon Cho Scientific Reports, 2026 Diabetic retinopathy (DR) is one of the major causes of preventable blindness in the world, and accurate large-scale screening tools are needed urgently. Most of the deep learning methods which have been developed for retinal image analysis are treating tasks like optic disc segmentation and DR grading separately. This separation is making it difficult for the model to use the shared anatomical and contextual cues which are linking the two tasks. So we are proposing GTAM-Net, a Gated Task-Attentive Multi-Task Network for retinal image analysis. GTAM-Net is performing optic disc segmentation and DR severity grading together inside a single end-to-end network. Inside the network, a gated task-attentive block is deciding how the features should be shared between the two tasks at each layer. In this way the network is keeping the useful complementary information for each task, and at the same time it is avoiding the negative transfer which often hurts multi-task models. We are also using a multi-scale feature pyramid for keeping the hierarchical context, and an uncertainty-based loss weighting so that one task is not dominating the training. The proposed method is tested on five public datasets: IDRiD, DDR, Messidor-2, APTOS, and REFUGE. The model is reaching up to 98.17% Dice score for optic disc segmentation and 99.12% accuracy for DR grading, and the performance of the proposed method is competitive on every dataset that we tried. The cross-dataset tests are also showing that the model is fairly stable when the imaging conditions are changing. From these results, the proposed multi-task design is appearing to be a useful and reasonably stable option for joint retinal image analysis, and it can be considered for use in large screening pipelines.
From Annotation to Prediction: Hospital-Grade Early Seizure Risk Prediction from Adult EEG Norah Alharbi, Mashael Aldayel, Shrooq Alsenan, Raneem Alyami, Enas Almowalad, Eman Alkethiry Diagnostics, 2026 Background: Manual review of EEG recordings in clinical settings is inherently time-consuming and labor-intensive. These challenges highlight a pressing need for automated EEG analysis tools capable of supporting clinicians by improving efficiency and diagnostic accuracy. Objectives: This study aims to develop and validate an AI-based model for the automated interpretation of adult EEG recordings. Unlike previous approaches that emphasize seizure detection during ictal states, our model targets the early prediction of seizure risk through systematic annotation and recognition of interictal patterns. Methods: The model is designed to accurately distinguish between normal and abnormal EEGs, encompassing both interictal and ictal activity. Abnormal EEGs will be further classified into three clinically relevant categories: (1) non-epileptiform abnormalities such as focal or diffuse slowing, (2) epileptiform discharges, and (3) electrographic seizures. Three AI-based classification algorithms were implemented: Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN). Results: RF demonstrated optimal performance across most tasks, achieving 96.50% accuracy for normal activity identification. This AI-driven system enhances the efficiency, consistency, and accessibility of EEG interpretation. It is particularly valuable in settings with limited access to neurophysiologists and offers an innovative approach to improving diagnostic timelines and clinical decision-making. Conclusions: Ultimately, this tool will support physicians in diagnosing neurological conditions and monitoring patient progress over time.
State estimator and convolutional neural networks-based fault localization approach for modern grids Jameel Ahmed Bhutto, Sohaib Tahir Chauhdary, Saad Arif, Akbar Niaz, Khaled Alnamasi, Syed Quadir Moinuddin, Shrooq Alsenan, Muhammad Attique Khan Scientific Reports, 2026 Direct current distribution networks get more attention due to the high penetration of renewable energy-based distributed generation. However, these networks face protection challenges like detecting, classifying, and localizing faults. This manuscript proposes a two-stage approach for identifying and isolating faults within these critical networks. The first stage uses a cubature Kalman filter to estimate the state of the measured current and voltage signals at the faulty bus. These estimated signals represent various fault scenarios and serve as convolutional neural network training data. Then, a total harmonic distortion index was computed from the estimated fault current for fault detection. Moreover, the direction of this reactive power is then utilized to classify the fault type and locate faulty sections in these networks. Extensive simulations demonstrate that the proposed scheme achieves fault detection under five milliseconds, exhibits 98% accuracy, and requires reasonable computational resources.
Audit-as-code: a policy-as-code framework for continuous AI assurance Aoun E. Muhammad, Kin-Choong Yow, Shrooq Alsenan Frontiers in Artificial Intelligence, 2026 Introduction Existing AI assurance and governance frameworks rely heavily on documented written policies and manual reviews of the implementation. The primary challenge is not the length of these documents, but to operationalize the gap from transforming qualitative requirements into verifiable controls. This approach makes ensuring continuous compliance through the development life cycle hard to enforce, scale, and reproduce. Methods This study presents a continuous assurance framework called Audit-as-Code that maps governance requirements to technically-auditable rules, that can be a combination of versioned policy specification and executable checks for evidence artifacts, linked to structured evidence regarding data, models, provenance, performance, decisions, and explanations regarding the decisions being made. While the framework addresses the governance and regulatory mapping requirements, the primary focus of this study is MLOps/CI-CD (continuous integration/continuous delivery) operationalization, and turning these requirements into deterministic checks and gate decisions integrated in operational workflows. In this study, we introduce an assured readiness score that integrates the governance risk with other key responsible AI principles, such as traceability and explainability. This approach helps in aligning deployment decisions with predefined risk tiers, and the framework automates decisions on whether a system can proceed, requires remediation and fixes, or should be blocked. It also provides targeted suggestions for improvement and compliance for the lags identified. Results We evaluated this framework on representative AI systems and demonstrated how a single evidence bundle can be used to support assessment across different governance regulations. Discussion In doing so, Audit-as-Code ensures AI assurance transforms from a documentation-driven policy module to a quantitative, auditable, reproducible, and operationally practical module to ensure compliance.
Morphology-guided attention networks for explainable skin cancer detection under clinical uncertainty Muhammad Zaheer Sajid, Muhammad Fareed Hamid, Zepa Yang, Mohammad Alhefdi, Shrooq Alsenan, Yunyoung Nam Frontiers in Oncology, 2026 Accurate and reliable skin cancer detection from dermoscopic images remains challenging due to large visual variability, overlapping lesion appearances, and inherent clinical uncertainty. To address these issues, this work proposes a morphology-guided attention framework for explainable and uncertainty-aware skin lesion classification. The system integrates lesion segmentation to preserve clinically meaningful morphological structures, followed by an attention-based classification network that emphasizes diagnostically relevant regions while suppressing background artifacts. Visual attention and attribution maps are generated to provide transparent explanations aligned with established dermoscopic criteria. In addition, an uncertainty estimation module is incorporated to quantify prediction confidence and identify ambiguous or out-of-distribution cases for safe clinical triage. The proposed approach is evaluated on publicly available dermoscopic datasets and achieves classification accuracy 99.12% with a recall rate above 99% for malignant lesions, demonstrating strong sensitivity for early cancer detection. Experimental results show that morphology-guided attention improves both classification performance and interpretability compared to conventional deep learning models. Furthermore, uncertainty-aware predictions enhance model reliability by reducing overconfident errors in challenging cases. These findings indicate that the proposed framework offers a robust, explainable, and clinically relevant solution for automated skin cancer screening under real-world conditions.
Enhancing transparency understanding using machine learning and visual analytics Samiha Fadloun, Sara Laouadi, Souham Meshoul, Kheireddine Choutri, Shrooq Alsenan, Mahmood Hosseini, Mohammed El Seddik Roukbi Peerj Computer Science, 2026 In today’s digital age, users are frequently confronted with lengthy terms and conditions documents associated with various products and services. Such documents often reference multiple entities (such as stakeholders, individuals, and users), with certain entities repeated throughout, underscoring their relative importance within the text. This study proposes a novel approach to facilitate the comprehension of terms and conditions by enhancing the detection and weighting of entities, as well as identifying relationships among them. By leveraging machine learning techniques (particularly natural language processing (NLP)) in conjunction with visual analytics, we aim to improve transparency and accessibility. Furthermore, we present an improved version of TranspVis, a visual analytics system to provide a more intuitive representation of transparency-related information. The proposed approach is evaluated through a combination of case studies and user experiments, offering a comprehensive assessment of its utility in rendering complex legal documents more interpretable. The findings underscore the potential of such tools to support large-scale applications in legal domains, with expert feedback affirming the value and relevance of the proposed solution.
A Bayesian Backed DeepLabV3+ Customized Hybrid-Depth Framework for Brain Tumor Segmentation and Classification Muhammad Sami Ullah, Muhammad Attique Khan, Saeed Iqbal, Shrooq Alsenan, Areej Alasiry, Mehrez Marzougui, Yunyoung Nam IEEE Access, 2026 Efficient diagnosis, treatment planning, and patient care requires that brain tumors in MRI images must be accurately segmented and classified. Many deep learning practices are used to perform these tasks. However, prominent performance gaps like insufficient hierarchical feature extraction, lesser-known biological delineation of tumor and limited uncertainty modeling exist. These gaps specifically occur while dealing with multi-modal MRI data and heterogenous tumor shapes. These gaps are addressed by introducing a Bayesian based DeepLabV3+ architecture. It quantifies aleatoric and epistemic uncertainty of segmentation. Hence, it provides improved reliability on clinical decisions. The selection of architecture is backed by the promising ability of Bayesian inference and preservation of multi-scale spatial context which is suited for complex morphology of brain tumors. The architecture is tested on BraTS23 and Figshare brain tumor datasets which has multi-modal annotations of tumor sub regions and consist of multiple tumor recognitions respectively. It performed remarkably to properly manage a range of tumor forms and sizes by achieving Dice scores of 95.86% on BraTS23 and 92.80% on the Figshare dataset for segmentation tasks, which its statistical significance is proven through several cross-validation trials (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">p</i> < 0.05) and superior performance compared with leading approaches reported in recent literature. Furthermore, a unique Hybrid-Depth classification framework is developed, which merges hierarchical and coarse-level feature extraction using dual-branch processing and self-attention. it is a design choice that is justified by the requirement to explicitly characterize cross-modal interdependence and mitigate feature redundancy. The framework performed effectively to classify brain tumor MRI modalities (Flair, T1, T2 and T1ce) and tumor type classes (i.e. gliomas, meningiomas and pituitary). The hierarchical and coarser feature extraction for classification task is done gracefully. The proposed architecture achieved 98.80% and 95.40% accuracy for the BraTS23 and Figshare brain tumor datasets, respectively. The work demonstrates its uniqueness and practical usefulness by defining the Hybrid-Depth mechanism and detailing its role in increasing discriminative representational learning. The framework’s performance demonstrates both its potential for practical use in automated brain tumor diagnosis and its superiority over current approaches. The proposed framework provided a major step toward increasing diagnostic efficiency, accuracy and assisting healthcare providers in providing the best possible patient care. In a nutshell, the proposed Bayesian-DeepLabV3+ and Hybrid-Depth frameworks jointly enhance the field by bridging methodological gaps in quantification of uncertainty, selection of dataset-driven architecture, and multi-modal feature learning. Hence, the approach offers a robust AI pipeline for automated brain tumor diagnosis.
LEAF: A Lightweight Language-Enhanced Model for Forestry Analysis in Remote Sensing Imagery Sanjar Karshiev, Faisal Saeed, Jaesin Ahn, Abdul Rehman, Muhammad Diyan, Shrooq Alsenan, Heechul Jung IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2026 Effective monitoring of deforestation requires accurate interpretation of satellite imagery, where domain expertise and computational efficiency are critical. General purpose vision language models (VLMs) perform well on natural images but struggle in remote sensing (RS), where spatial structure and fine-grained anthropogenic cues (e.g., artisanal mining, selective logging) are underrepresented in web-scale corpora. We introduce LEAF (Language-Efficient Analysis of Forestry), a lightweight, instruction-tuned small-scale vision language model (SVLM) fine-tuned on our proposed DForest-VL dataset, containing 1,500 image-text conversational pairs, spanning image-, region-, and grounded-level tasks. Our SVLM achieves strong multilabel classification (Recall = 0.93, F1 = 0.92, Hamming Loss = 0.048), outperforming alternative pairings in ablations and matching or surpassing supervised vision backbones trained on the full dataset. On deforestation-specific categories, LEAF attains the lowest average Hamming Loss (0.038) across methods, with salient gains for dispersed patterns such as selective logging. Qualitative analysis on remote sensing visual question-answering (RS-VQA) further shows reduced hallucination, mis-localization, and generic responses relative to large VLMs (e.g., LLaVA, GeoChat, GPT-4). With a sub-3B parameter footprint and single-GPU training, LEAF lowers the barrier for deployable, domain-specific VLMs. These results highlight the data efficiency, accuracy, and practical utility of targeted instruction tuning for real-world deforestation monitoring.
A novel deep semantic- and vision-based self-attention architecture for skin cancer classification Junaid Aftab, Muhammad Attique Khan, Sobia Arshad, Amir Hussain, Shrooq Alsenan, Yongwon Cho, Yunyoung Nam Digital Health, 2026 Objectives In the world, skin cancer is a significant health concern, and early diagnosis of this cancer plays a key role in improving patient outcomes. The early detection of this cancer reduces the death rate, but due to the complexity of the diagnosis, incorrect detection and prediction are provided by the experts. Therefore, it is essential to propose a computer-aided diagnostic system based on deep learning and explainable Artificial Intelligence (XAI) techniques that can be used as a second opinion in clinics and help physicians more accurately detect and predict this type of cancer. Methods This work presents the proposed deep learning architecture consisting of two modules—skin lesion segmentation and lesion type classification. The proposed architecture is interpreted using XAI techniques to better evaluate the black-box model. In the skin lesion segmentation phase, we implemented DeepLab V3 architecture for semantic segmentation. The ResNet-18 model was used as the backbone, and later hyperparameters were optimized using Bayesian Optimization (BO). In the classification phase, we design a FusedNet architecture called Inverted self-attention with Vision Transformer (ISAwViT). The proposed fused network combines an inverted self-attention residual architecture with a vision transformer. The proposed fused network extracted feature information more deeply than performing an accurate prediction in a later stage. The design model is trained, and later in the testing phase, extracted features are classified using Softmax and several other classifiers. Results The lesion segmentation and classification experiment was conducted on the HAM10000 dataset. The accuracy achieved by the HAM10000 dataset was 95.16% for lesion segmentation and 97.5% for lesion classification. Conclusion Compared with recent techniques, the proposed model is more effective and efficient. In addition, the interpretation of the proposed model was performed using LIME and Grad-CAM, which show how the fused model makes correct classifications.
SDMCC: Sample-wise Debiased Multilevel Contrastive Clustering for Single-cell Gene Expression Data Han Xiao, Dayu Hu, Fengyue Zhang, Shrooq Alsenan, Por Lip Yee, Jing Yang, Xiaoyu Cui IEEE Journal of Biomedical and Health Informatics, 2026 Single-cell gene expression profiling has emerged as a powerful technology for dissecting complex tissues at unprecedented resolution. Accurate cell clustering is a fundamental computational prerequisite for cell type identification. In recent years, numerous single-cell contrastive clustering algorithms have been developed to more effectively identify individual cells and characterize cellular heterogeneity. Despite these advances, the presence of noise and data sparsity in single-cell datasets continues to impede algorithmic performance. Mainstream clustering methods face three principal limitations: (i) most clustering methods process data in mini-batches, and batch-related value shifts can destabilize feature extraction; (ii) existing contrastive learning strategies fail to achieve multilevel, coordinated information fusion, making them prone to collapsed solutions; and (iii) noise and sparsity can cause centroid shifts, yet existing methods lack effective mechanisms for dynamic centroid updating. To address these challenges, we propose sample-wise debiased multilevel contrastive clustering with shrinkage risk regularization (SDMCC). Instead of directly learning from the original sparse data, SDMCC applies a sample-wise correction module in conjunction with a batch-size adaptation module to mitigate distortion. Furthermore, we introduce a multilevel contrastive learning strategy that integrates instance-level and cluster-level information, enabling the capture of fine-grained cellular variations while maintaining coarse-grained semantic consistency across cell subpopulations. In addition, a shrinkage risk regularization term prevents convergence to trivial solutions. Extensive experiments conducted on multiple single-cell datasets demonstrate that SDMCC not only achieves superior clustering accuracy but also effectively uncovers biologically meaningful patterns in gene expression, thereby highlighting its potential for broader biomedical applications.
Deep Learning Reader for Visually Impaired Jothi Ganesan, Ahmad Taher Azar, Shrooq Alsenan, Nashwa Ahmad Kamal, Basit Qureshi, Aboul Ella Hassanien Electronics Switzerland, 2022
