cv

Master of Applied Science in Electrical Engineering

• Thesis Work (In progress).
  • Built a novel controlled video diffusion model to generate synthetic ultrasound videos of the heart
  • Achieved state-of-the-art ejection fraction estimation performance using synthetic data.

Bachelor of Applied Science in Eelectrical Engineering

Machine Learning Engineer

• Created a similarity search microservice capable of producing a list of 100 similar in 30 milliseconds.
• Enhanced similarity search microservices, resulting capacity increase to 100,000 files.
• Transitioned a large-scale Elasticsearch service to PostgreSQL, yielding a 5x increase in asset handling and a 25% boost in processing speed.
• Reduced AWS-managed services costs by 30% through system optimization and monitoring usage.

Graduate Student Researcher

• Built a novel controlled video diffusion model to generate synthetic ultrasound videos of the heart conditioned on the patient’s other acquired views.
• Achieved state-of-the-art ejection fraction estimation performance using synthetic data.

Undergraduate Researcher

• Launched an Android mobile application for real-time heart ultrasound image analysis, achieving a 30% increase in processing speed and user experience.
• Integrated the application with third-party tools, enhancing data visualization and patient engagement.

Multi-Tasking Transformer

• Engineered a ViT model for cardiac cycle classification improving prediction accuracy by 5%.
• Achieved a 5% increase R2 score in EF prediction accuracy compared to baseline.
• Achieved maximum accuracy of 85% in classifying the frame label using the classification head.
• Explored the addition of extra task-specific class tokens to improve the accuracy of the tasks concurrently.

Representation Learning Using GANs

• Collaborated in developing a generative adversarial network (GAN) as a video-to-image encoder to improve the accuracy of ejection fraction estimation of heart Echocardiograms.
• Implemented a pix2pix-based GAN network to generate synthetic images from ultrasound videos of the heart that capture the shape of the left ventricle accurately.
• Achieved better representation learning measured by the performance of the downstream tasks.
• Achieved higher convergence speed by up to 30% along with higher accuracy in ejection fraction as measured by metrics such as MAE and F1 score

Estimating EF Using GNNs and Contrastive Learning

• Implemented a contrastive loss function in a(EchoGNN) to improve ejection fraction prediction accuracy in estimating heart’s EF.
• Conducted extensive ablation experiments to compare the performance of the model with and without the contrastive loss.
• Achieved a notable increase in ejection fraction prediction accuracy compared to baseline, as measured by metrics such as F1 score and Mean Absolute Error (MAE).

• Graduate Support Initiative (GSI)

• The Canada Graduate Scholarship (CGS-M)
• Neekoo Grant Award

• Undergratuate Research Student Award (NSERC URSA)
• IPex Innovation Scholarship

Academic Services

• Reviewer for NeurIPS AFM workshop
• Reviewer for IEEE Transactions on Medical Imaging (TMI)