AI Researcher
Computer Vision • Medical Imaging • Explainable AI
I am interested in developing reliable and explainable AI systems for medical image analysis, with a focus on disease detection, clinical decision support, and trustworthy deep learning models.
Peer-reviewed papers and ongoing manuscripts
End-to-end AI systems and research implementations
A hybrid deep learning framework combining ResNet-based feature extraction with Vision Transformer attention mechanisms for automated tuberculosis screening from chest X-ray images.
A systematic study of data augmentation pipelines for pneumonia classification from chest X-rays, targeting performance improvement under limited data conditions.
A real-time API service for detecting emergency events (falls, fights, intrusions) from security camera feeds using YOLOv8 with low-latency inference.
My long-term research vision is to develop AI systems that physicians can genuinely trust and use in clinical practice. Current deep learning models often achieve high accuracy on benchmark datasets but fail to generalize robustly or explain their decisions in ways clinicians can interpret.
I am motivated by three core research questions: How can we build hybrid architectures—combining the local feature extraction of CNNs with the global context modeling of Vision Transformers—that perform reliably even under limited labeled data? How can explainability methods be meaningfully evaluated so that saliency maps actually correspond to clinically relevant regions? And how can multimodal learning across imaging modalities improve diagnostic accuracy for diseases where single-modality data is insufficient?
During my PhD, I intend to pursue these questions in the context of chest X-ray analysis, ultrasound imaging, and potentially retinal imaging—domains where AI has clear potential to augment clinical workflows and improve patient outcomes in resource-limited settings.
Research notes, paper reviews, and reflections
A practical account of designing, training, and evaluating a hybrid CNN-Vision Transformer architecture for tuberculosis detection — including the mistakes, trade-offs, and insights along the way.
Why saliency maps are not enough, what clinicians actually need from AI explanations, and how researchers should approach XAI evaluation in the medical domain.
A review of quantitative and qualitative evaluation frameworks for saliency maps, including pointing game, insertion/deletion metrics, and radiologist agreement protocols.