New data augmentation methods, refined priors, and a more robust loss function lead to unprecedented (Dec 2023) performance on a 3D CNN for image segmentation. Project culminated in the algorithm being implemented in a real medical setting and a paper accepted at IEEE ISBI 2024
Developed as part of my Master's thesis at Harvard (in collaboration with MIT and MGH), this project was used to leverage WMH-SynthSeg, an extension of SynthSeg, designed for brain MRI scans. It was implemented in a medical setting since it is able to uniquely adapts to low-field MRI scans, providing high-resolution segmentations of white matter hyperintensities (WMH) and anatomical structures, even in challenging imaging conditions.
Author: Pablo Laso
Email: plaso@kth.se
Citation: If you use WMH-SynthSeg in your analysis, please cite our paper under review.
- Quantifying white matter hyperintensity and brain volumes in heterogeneous clinical and low-field portable MRI. Master's thesis at MGH, Harvard. View Project
- AI-based Prostate MRI Detection at GE: An advanced CAD system for detecting and classifying prostate cancer from MRI scans. View Project
- Deep Learning for Lung Ultrasound Imaging: Utilized CNNs and transfer learning to identify pneumonia and COVID-19 from LUS images. View Project
- Gest2talk: A myo-armband project for aiding individuals with speech impairments communicate through gesture recognition. View Project
- π Education: M.Sc. in Computer Science from the University of Twente (Netherlands), with previous studies at KTH (Sweden) in Data Science and Statistics. Master's thesis on Deep Learning, as a student at the MGH-Martinos Center (MIT and Harvard Medical School), Boston, MA, USA.
- π¨βπ» Experience: Former Engineer Research Assistant at Karolinska Institutet (Sweden), where I enhanced image processing workflows and data management for medical research. Previous work at GE, on AI algorithm development.
- π Location: Boston, MA, working on AI applications in healthcare.