We take pride in our talented team of engineers who are constantly pushing the boundaries of what is possible in data science. Today, we are incredibly proud to spotlight Samuel, one of our brilliant computer scientists, who is currently completing a groundbreaking master's project at the University of Stavanger (UiS), that aligns with our mission to shape the future of data.

Reconstructing the Brain

Samuel’s master's thesis is focused on generating synthetic “pre-stroke” brain MRI images using deep learning. Given that many stroke patients do not have scans prior to their event, there is a gap in our understanding of their prior brain health. And without this "before" picture, planning the most effective rehabilitation is a significant challenge.

To address this issue, Samuel is developing a 3D reconstruction model that uses deep learning to generate synthetic "pre-stroke" brain images. By learning the complex anatomical structures of healthy brains, his model can reconstruct entire missing regions, starting with an entire hemisphere to provide a foundation for estimating brain health that isn't biased by stroke damage.

This project was one of several options within his MSc program at the University of Stavanger (UiS), and it resonated deeply with Samuel as it combines machine learning with tangible medical impact.

“I’ve always been interested in Machine Learning, and knowing that this work could contribute to stroke research and rehabilitation planning made it especially meaningful,” Samuel shares.

By successfully implementing this approach, Samuel hopes to enable clinicians to make more accurate assessments of a patient’s underlying brain health, improving rehabilitation strategies and long-term outcome predictions.

Solving the "Missing Data" Problem in Healthcare

The main objective of Samuel's research is to construct a robust 3D deep learning model capable of reconstructing missing brain tissue in MRI scans.

The solution works by taking scans of healthy brains and hiding one half. The AI then tries to rebuild the missing side by looking at the symmetry of the visible half. This allows him to measure exactly how accurate the model is before he begins testing it on brains damaged by strokes.

Innovation rarely comes without obstacles, and Samuel has also faced a few challenges throughout his project, including:

• The absence of true ground truth pre-stroke images.
• The complexity and individuality of brain anatomy.
• The distortion of MRI scans by stroke lesions, which can skew traditional brain health metrics.

The complexity of Samuel’s work is a testament to the high-level engineering we value at IntelliStream.

Work in Progress & Future Impact

Samuel has already implemented the full end-to-end pipeline, which includes preprocessing, image masking, reconstruction training and validating his large scale training on several hundred subjects.

“I’m currently refining the reconstruction strategy, exploring improved mirroring techniques and anatomical constraints to enhance anatomical realism”. - Samuel

Looking ahead, Samuel’s approach holds the promise of generating synthetic pre-stroke images for patients who only have post-stroke scans. This advancement could lead to more accurate estimations of brain health and ultimately support improved rehabilitation planning. Moreover, his work contributes to the expanding field of generative AI in medical imaging, signaling the potential for significant advances in healthcare technology.

From Research to Real-World Application

While Samuel's current focus is medical imaging, the skills he is mastering have a direct impact on our work at IntelliStream.

“This project has strengthened my ability to design large-scale ML pipelines, work with high-dimensional 3D data, and optimize GPU-based systems,” says Samuel. “It has also improved my ability to structure complex problems and communicate technical concepts clearly, skills that transfer directly into my work at IntelliStream.”

Samuel's innovative work is a prime example of the IntelliStream team's commitment to leadership in the future of data. We are excited to witness the continued advancement of medical imaging and data science through his dedicated use of data for real-world impact.

Stay tuned as we share more updates on our talented team and their remarkable projects!