Haris Ghafoor

MS-PhD in Computational Science and Engineering at Yonsei University, Korea

About Me

Haris Ghafoor

Computational Science Researcher

I am a Computational Science and Engineering researcher focused on inverse learning problem modeling, ultrasound synthesis methods, and computational methods for medical imaging.

With a background in Electrical Engineering and expertise in Digital Signal Processing and Machine Learning, I bring a multidisciplinary approach to solving complex computational problems in medical imaging and diagnostics.

Machine Learning
Medical Imaging
Signal Processing
Computational Methods
Inverse Problems

Work Experience

Graduate Researcher

Computational Methods in Medical Engineering Lab (CMME)

Oct 2024 - Present Yonsei University, Seoul, South Korea

Working on developing advanced medical image processing methods for segmentation and enhancement using AI-inspired computational approaches.

AI Team Lead

eMoldino

Nov 2023 - Oct 2024 Seoul, South Korea

Led a 5-member team to develop AI-powered tooling monitoring solutions for supply chain optimization. Developed AI models for anomaly detection in production cycles using statistical rules, deep contrastive learning, and autoencoders for robust & accurate quality detection.

AI/ML Freelancer

Upwork

Present Remote

Providing freelance consulting services in machine learning and AI development. Specialized in time series forecasting, computer vision, and deep learning solutions for various clients globally.

AI Researcher

Deep Chain Solutions

Mar 2022 - Nov 2023 Remote, Seoul, South Korea

Developed global forecasting models for time series, using multitasking techniques for robust performance. Utilized pattern recognition and classification methods to identify production phases in tool-making machine data.

Research Focus

Exploring computational methods for medical imaging and ultrasound synthesis.

Inverse Learning Problems

Developing mathematical models to solve inverse problems in medical imaging.

Key Focus Areas

  • Regularization techniques
  • Uncertainty quantification
  • Bayesian approaches
  • Deep learning integration

Ultrasound Synthesis

Creating novel methods for ultrasound image synthesis and enhancement.

Key Focus Areas

  • Generative models
  • Image-to-image translation
  • Speckle noise reduction
  • Resolution enhancement

Computational Methods

Advancing algorithmic approaches for medical image analysis.

Key Focus Areas

  • Fast iterative methods
  • Sparse reconstruction
  • GPU-accelerated computing
  • Real-time processing

Featured Projects

PhyUSFormer

Physics-Guided Ultrasound Segmentation Transformer that leverages a physics-guided simulation pipeline to generate synthetic datasets for foundation model pretraining.

Deep Learning Physics-Based Modeling Medical Imaging
90.57% Dice on BUSI
89.82% Dice on UDIAT

GDSSA-Net

A gradually deeply supervised self-ensemble attention network for IoMT-integrated thyroid nodule segmentation in ultrasound images.

IoMT Deep Supervision Self-Ensemble
79.85% Dice on DDTI
84.27% Dice on TN3K

Electricity Demand Forecasting with LSTM-NBEATS

Development of a Network of Stacked Blocks of MLP in which each block is encoded by Recurrent Layer i.e. LSTM in order to accurately capture the different kinds of trends & seasonality in Time Series.

LSTM Time Series Forecasting Deep Learning
Enhanced pattern recognition
Improved forecasting accuracy

Publications

2023

GDSSA-Net: A gradually deeply supervised self-ensemble attention network for IoMT-integrated thyroid nodule segmentation

Muhammad Umar Farooq, Haris Ghafoor, Azka Rehman, Muhammad Usman, Dong-Kyu Chae

Internet of Things

The integration of deep learning techniques in the Internet of Medical Things (IoMT) has significantly advanced the early detection of life-threatening diseases such as thyroid cancer. Our model achieves a Dice coefficient of 79.85% and 84.27% on DDTI and TN3K datasets, respectively.

2024

PhyUSFormer: Physics-Guided Ultrasound Deep Learning Net for Image Segmentation

Haris Ghafoor, et al.

Under Submission

A framework that leverages the physics of wave propagation in tissue to supervise the training of a deep learning model for robust ultrasound segmentation. Evaluations on public breast ultrasound datasets achieved significant improvements, with Dice scores of 90.57% on BUSI and 89.82% on UDIAT.

Contact Me

Email

ghafoorharis@yonsei.ac.kr

Location

Yonsei University, Seoul, South Korea

Connect With Me

Social Profiles