My research is broadly organized into six domains, shaped by industrial needs and international collaborations.

In Biomedical, the primary focus is on AI-driven clinical decision support systems that assist in medical interpretation and diagnosis. A key emphasis is on leveraging multimodal healthcare data, including X-ray, MRI, and sensor-based signals, to build robust predictive and assistive models. These systems aim to enhance diagnostic accuracy, support early detection of diseases, and improve clinical decision-making by integrating imaging and physiological information in a unified analytical framework. This domain represents one of the central pillars of my research, with strong emphasis on real-world clinical impact.

In Connected Vehicles, the work is extensively focused on intelligent transportation systems enabled by AI and real-time sensor analytics. This includes processing data from onboard vehicle sensors and external infrastructure to understand road environments, predict traffic behavior, and enhance situational awareness. The objective is to improve road safety, enable adaptive and intelligent driving support, and contribute to the development of autonomous and connected mobility ecosystems. This is another core research area with broad industrial relevance and active collaboration.

In Energy, the work focuses on data-driven forecasting and optimization of energy systems to improve efficiency and support sustainable energy management.

In Waste Management, the emphasis is on smart, data-enabled approaches for efficient waste collection, classification, and process optimization to support sustainable urban systems.

Materials and Multiphase Systems represent additional miscellaneous top research areas that I have been working on. In Materials, the focus is on modeling and predicting material behavior to aid in the design of more reliable engineering components. In Multiphase Systems, the work focuses on modeling and understanding interactions between solids, liquids, and gases in complex industrial processes to improve efficiency and performance.

Together, these domains reflect the application of AI and engineering methods to complex real-world industrial challenges. Some of my key works on the individual domain are given below for exploration.

Biomedical Domain

In the Biomedical domain, my research spans multiple critical areas aimed at advancing medical diagnostics and treatment. My work includes developing AI-driven clinical decision support systems for conditions such as Handostenonet, Skin Cancer, Gastrointestinal disorders, Diabetic Sensorimotor Polyneuropathy (DSPN), and Alzheimer's Disease. Additionally, I focus on EEG reconstruction techniques and EMG-based prosthetic control through NeuroAttend-EMG, enhancing assistive technologies for patients with prosthetic needs. These projects involve leveraging multimodal healthcare data, including imaging, sensor signals, and neural data, to build predictive models and improve clinical decision-making. The attached works further illustrate the innovative applications and potential impact of this research on improving patient outcomes and healthcare practices.

Handostenonet

We have developed HandOsteoNet, an AI-driven framework for automated bone age assessment aimed at improving pediatric healthcare. By integrating a RegNetY-400MF encoder with a Swin Transformer decoder, dual-path feature extraction, and graph convolutional networks, the model achieves a mean absolute error of 3.86 ± 0.37 months, significantly outperforming traditional methods. Preprocessing with CLAHE enhances feature representation from hand X-rays, improving accuracy by up to 26.8%. With real-time processing and gender-aware predictions, HandOsteoNet demonstrates robust generalization across external datasets, making it ready for clinical deployment. A web application integrates the system for real-time prediction and validation, with Grad-CAM visualizations ensuring clinical interpretability. The current paper detailing this work is under review.

DSPN

We are currently working on a non-invasive, non-subjective platform for rapid DSPN screening, utilizing a portable thermography system with 50 high-resolution infrared sensors. This project addresses the growing diabetes burden in regions like Qatar, aiming to reduce DSPN-related healthcare costs by over 80%. The system captures plantar thermogram images, and a convolutional neural network (CNN) model, trained on data from Bangladesh, Malaysia, and Qatar, classifies DSPN severity with 90% accuracy. The approach includes multi-center data, Grad-CAM visualizations for enhanced interpretability, and aims to improve early detection in resource-limited settings. Additional papers related to this work are attached below to explore.

Paper Links:

Machine learning model for stratification of early diabetic foot complication using thermogram images

Skin Cancer

We worked on Efficient-DWMF-Net, an intelligent multimodal skin cancer classification framework designed to improve the detection of skin lesions by combining clinical images, dermoscopic images, and patient metadata. The model uses EfficientNetB0, skip-gate components, and CBAM attention mechanisms to strengthen feature extraction while maintaining computational efficiency. A key contribution of this work is the use of a dual-stage weighted fusion strategy, where different modalities are combined based on their diagnostic contribution, enabling stronger multiclass and multitask classification performance. The results show that carefully designed multimodal fusion improves accuracy, precision, recall, specificity, and F1-score across multiple classification tasks, making the framework a promising tool for supporting dermatological diagnosis. The current paper of this work is under review.

Gastro

We worked on HierTAC-Net (Hierarchical Triple Attention ConvNeXt Network), an explainable deep learning model for automated gastrointestinal (GI) endoscopic image classification, aimed at improving colorectal cancer detection. The model integrates Inverted Residual Blocks, Triple Attention modules, and ConvNeXt-based Feature Refinement Blocks to provide flexible classification across seven configurations, from binary polyp detection to full 8-class lesion classification. Trained on Kvasir-v2 and externally validated on HyperKvasir, HierTAC-Net achieved 94.21% F1-score on Kvasir-v2 and 96.03% on HyperKvasir, outperforming all compared models. The model demonstrates robustness under blur conditions and operates significantly faster than expert gastroenterologists, with diagnostic accuracy exceeding that of an 8-year expert. This research offers a clinically validated, efficient, and interpretable solution for real-time GI abnormality classification. Additional papers related to this work are attached for further exploration.

Paper Links:

Alzheimer

We worked on an explainable dual-head multi-scale 1D residual CNN-KAN framework for Alzheimer’s disease diagnosis and severity estimation, aimed at bridging the gap between diagnosis and progression assessment. This model uses over 200,000 clinical records from the National Alzheimer’s Coordinating Center, integrating cognitive scores, neuroimaging markers, genetic data, and health indicators. It achieves 91.34% classification accuracy, an F1-score of 91.16%, and an R² of 0.81, demonstrating robust generalization. The framework also identifies key predictive features, such as vascular and metabolic indicators (blood pressure, diabetes), which could become modifiable targets for future interventions. By unifying disease classification, progression estimation, and feature interpretation within a SHAP-based explainable model, this work supports clinical decision-making and offers a more transparent AI solution for Alzheimer's diagnosis. The current paper on this work is under review.

I have also attached additional papers related to my work in the Biomedical domain, focusing on EEG reconstruction and EMG-based prosthetics control, which explore innovative approaches for neural signal processing and assistive technologies. These papers provide further insights into the development and application of AI-driven solutions in clinical and prosthetic systems for improved patient outcomes.

Paper links:

Connected Vehicle Domain

I have extensively researched Connected Vehicle Systems, focusing on utilizing smartphone and onboard sensors combined with machine learning for real-time, privacy-preserving road condition and driving behavior monitoring. My work leverages federated learning and multisensor data (accelerometer, gyroscope, GPS, weather, GIS) to detect road anomalies without compromising data privacy. Some of my notable contributions include "Harnessing Smartphone Sensors for Enhanced Road Safety", "Explainable Attention-Augmented LSTM Model for Robust Real-Time Detection of Driving Behavior Patterns and Road Anomalies", "Federated Learning Approach for Scalable Road Condition Detection", "A Multimodal Smartphone & Camera Dataset for Holistic Road-Way Analysis", and "IEEE Digital Privacy Model for Connected Vehicles and Intelligent Transportation Systems". The papers related to these works are attached below for exploration.

Paper links:

Waste Management Domain

I have also worked in the Waste Management domain, focusing on AI-based and sustainable approaches to improve waste classification, sorting, and recycling. My research includes developing deep learning models for accurate waste classification and real-time detection, along with automated systems designed for efficient sorting and recycling. Additionally, my work highlights the use of organic waste compost to enhance soil fertility, effectively linking waste management practices with sustainable agriculture. Some of my notable contributions in this domain are attached for exploration.

Paper links:

Energy Domain

In the Energy domain, I have focused on improving energy system performance in challenging and dynamic environments. My research combines adaptive machine learning for photovoltaic (PV) power prediction with advanced control strategies to enhance microgrid stability. This includes the development of coating-aware, incremental learning models to improve PV yield forecasting under desert dust conditions and the use of electric vehicle-based virtual inertia control to mitigate frequency fluctuations in renewable-rich islanded microgrids. Some of my key works in this domain are attached below for further exploration.

Paper links:

Material Domain

In the Materials domain, my research focuses on the performance of thin metal and semiconductor films, which are essential in optoelectronic and sensing applications. I explore how factors such as composition, thickness, and interfacial properties influence their behavior. By combining machine learning with physics-based simulations, my work aims to predict and optimize optical performance and device efficiency, particularly for photodetectors and light-harvesting technologies. Some of my significant contributions in this domain are attached below for further exploration.

Paper links:

Innovative practices in Education Domain

In the Innovative Practices in Education domain, I have explored the integration of AI in engineering education, focusing on applications like intelligent tutoring systems, adaptive assessments, and VR/AR learning environments. My research also delves into the emerging use of generative AI and large language models to enhance learning experiences. The findings from these studies show moderate improvements in learning outcomes and student engagement, while also addressing concerns related to ethics, equity, and the long-term effectiveness of these technologies. Some of my notable works in this domain are attached below for further exploration.

Paper links: