Projects
Featured Projects
🧠 Early Alzheimer’s Detection via Multimodal Retinal Imaging
Key Innovation: Novel texture biomarkers in hyperspectral retinal imaging, multimodal classifiaction achieving AUC 0.84
- Developed end-to-end pipeline from image quality assessment to multimodal classification
- Created novel hyperspecrtal imaging texture biomarker for early disease detection
- Assessed repeatability and reproducibility of imaging biomarkers
Abstract (Preliminary results - full paper coming soon!)
👁️ Vision Foundation Models for Glaucoma Classification
Key Innovation: Addressing domain shift across imaging devices and demographics
UMAP embeddings of image features by dataset and classification label (Normal vs Glaucoma), showing domain shift between different imaging devices/clinical settings/demographics
- Leveraged Vision Foundation Models (VFM) pre-trained with self-supervised learning
- Implemented multi-source domain fine-tuning techniques
- Applied domain adaptation to handle demographic and device variations
- Achieved robust performance across different clinical settings
🐔 Camera-based Activity Recognition for Animal Welfare
Key Innovation: Hybrid approach combining object detection with motion analysis
Overview of the developed hybrid algorithm combining YOLOv5 object detection with motion analysis
Real-time head detection of laying hens using fine-tuned YOLOv5 on custom labeled dataset
- Developed real-time animal behavior monitoring system
- Combined YOLOv5 object detection with Gaussian Mixture Model motion detection
- Achieved early detection of welfare-compromising behaviors
Resources:
🧠 Multimodal Seizure Detection
Key Innovation: Fusion of multiple wearable sensor modalities for improved detection
- Modality specific feature extraction + SVM classification
- By combining multiple modalities, we could improve sensitivity and reduce false alarm rate
